CN118297245B - Energy consumption prediction method, device, computer equipment and storage medium - Google Patents

Abstract

The application relates to an energy consumption prediction method, an energy consumption prediction device, computer equipment and a storage medium. The method comprises the following steps: acquiring characteristic production data corresponding to at least two consumption influence characteristics of the energy to be predicted; determining a target usage prediction model matched with the feature scheduling data from at least two usage prediction models obtained by pre-training based on the feature scheduling data; and according to the characteristic scheduling data and the target consumption prediction model, predicting the consumption of the energy to be predicted, and obtaining the predicted energy consumption of the energy to be predicted. By adopting the method, the energy consumption prediction accuracy can be effectively improved.

Description

Energy consumption prediction method, device, computer equipment and storage medium

Technical Field

The application relates to the technical field of intelligent energy management and control, in particular to an energy consumption prediction method, an energy consumption prediction device, computer equipment, a storage medium and a computer program product.

Background

Along with the continuous rising of global energy prices, energy consumption management gradually becomes a core problem concerned by a plurality of enterprises in business production, and energy consumption in business production can be reduced and energy utilization efficiency can be improved by effectively managing energy.

At present, when business production is carried out, the energy consumption prediction is usually carried out based on methods such as time sequence analysis or regression analysis, and the accuracy of the methods is low when the energy consumption prediction is carried out aiming at processing a large amount of complex data, so that the current energy consumption prediction requirement cannot be met.

Disclosure of Invention

In view of the foregoing, it is desirable to provide an energy consumption prediction method, apparatus, computer device, computer-readable storage medium, and computer program product that can effectively improve the accuracy of energy consumption prediction.

In a first aspect, the present application provides a method for energy usage prediction, the method comprising: acquiring characteristic production data corresponding to at least two consumption influence characteristics of the energy to be predicted; for each consumption influence characteristic, acquiring the energy consumption correlation of the consumption influence characteristic and the energy to be predicted; determining key impact features from each of the energy usage impact features based on each of the energy usage correlations; determining a target usage prediction model from at least two usage prediction models according to the key influence features and feature scheduling data of the key influence features in the feature scheduling data; the at least two consumption prediction models of the energy to be predicted are respectively used for representing different energy consumption trends of the energy to be predicted in the production process; and according to the characteristic scheduling data and the target consumption prediction model, predicting the consumption of the energy to be predicted to obtain the predicted energy consumption of the energy to be predicted.

In the above embodiment, when the energy consumption prediction is performed, the feature scheduling data corresponding to each consumption influencing feature influencing the energy consumption of the energy to be predicted can be obtained, the business production situation when the energy to be predicted is used for production reflected by the feature scheduling data corresponding to each consumption influencing feature, is determined from at least two consumption prediction models obtained by training in advance, the target consumption prediction model matched with the feature scheduling data can be used, the target consumption prediction model is more suitable for the actual production situation when the energy to be predicted is used for production, and the consumption prediction is performed on the energy to be predicted according to each feature scheduling data and the target consumption prediction model, so that more accurate predicted energy consumption can be obtained, and the accuracy of the energy consumption prediction on the energy to be predicted is effectively improved. In addition, by determining key influence characteristics of the energy to be predicted and determining the target usage prediction model based on the key influence characteristics, the number of the finally determined target usage prediction models can be unique, and the determination efficiency of the target usage prediction model is effectively improved.

In one embodiment, the usage prediction model includes a predictor model corresponding to each of the usage influence features, and model types of the predictor models corresponding to each of the usage influence features included in the usage prediction model are the same; the determining the target usage prediction model from at least two usage prediction models according to the key influence feature and the feature production data of the key influence feature in the feature production data comprises: determining a target prediction sub-model from at least two prediction sub-models corresponding to the key influence features according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data; and determining a usage prediction model containing the target predictor model as the target usage prediction model.

In the above embodiment, at least two predictor models corresponding to the key influence features are determined first, and then the target predictor model is determined from the predictor models corresponding to the key influence features directly according to the key influence features and the feature scheduling data of the key influence features, so that the usage prediction model including the target predictor model can be determined as the target usage prediction model, thereby effectively improving the determination efficiency of the target usage prediction model and further improving the prediction efficiency of energy usage prediction.

In one embodiment, the usage prediction model includes a first usage prediction model having energy prediction vertex values; the determining a target prediction sub-model according to the key influence feature and the feature scheduling data of the key influence feature in the feature scheduling data from at least two prediction sub-models corresponding to the key influence feature includes: determining model selection data of the energy to be predicted according to an energy prediction vertex value of a predictor model corresponding to the key influence feature in the first quantity prediction model; and determining a predictor model corresponding to the key influence feature in the first quantity prediction model as the target predictor model under the condition that the feature production data of the key influence feature meets the use condition of the first quantity prediction model based on the model selection data.

In the above embodiment, the model selection data of the energy to be predicted is determined by the energy prediction vertex value of the predictor model corresponding to the key influence feature, and whether the feature scheduling data of the key influence feature meets the use condition of the first quantity prediction model is firstly determined based on the model selection data, so that whether the current production condition of the energy to be predicted is matched with the first quantity prediction model can be quickly determined, the first quantity prediction model is directly used under the matched condition, and the accuracy of model determination can be provided while the model determination efficiency is improved.

In one embodiment, the usage prediction model further includes a second usage prediction model, where the energy usage prediction value in the second usage prediction model is positively correlated with the feature production data of the usage-affecting feature; the method further comprises the steps of: and determining a predictor model corresponding to the key influence feature in the second usage prediction model as a target predictor model under the condition that the feature production data of the key influence feature is determined not to meet the usage condition of the first usage prediction model based on the model selection data.

In the above embodiment, under the condition that it is determined that the energy consumption prediction of the energy to be predicted by using the first consumption prediction model does not conform to the actual production rule, the energy consumption prediction is performed for the energy to be predicted by using the second consumption prediction model in which the energy consumption prediction value and the characteristic production data of the consumption influence characteristic are always in positive correlation, so that the energy consumption prediction process of the energy to be predicted conforms to the actual production rule more effectively, and the accuracy of the energy consumption prediction is improved.

In one embodiment, when the usage affecting feature includes an equipment start variable, the method for obtaining feature production data corresponding to the equipment start variable includes: acquiring related scheduling data of the device related features based on the device related features meeting the correlation conditions with the device opening variable; and processing the related production scheduling data according to the energy to be predicted and the equipment prediction model matched with the equipment opening variable to obtain the characteristic production scheduling data corresponding to the equipment opening variable.

In the above embodiment, for the equipment start variable that cannot directly obtain the feature scheduling data, the feature scheduling data of the equipment related features having strong correlation with the equipment start variable may be used, and the feature scheduling data of the equipment start variable is obtained by combining with the equipment prediction model obtained by pre-training, so that the consideration dimension of the usage influencing feature when the energy usage is predicted for the energy to be predicted can be effectively increased, and the accuracy of the energy usage prediction is further improved.

In one embodiment, the training mode of each usage prediction model includes: for each consumption influence characteristic of the energy to be predicted, acquiring historical characteristic data of each consumption influence characteristic and historical energy consumption data of the energy to be predicted; according to the historical characteristic data and the historical energy consumption data, constructing a training data set corresponding to each consumption prediction model; and respectively carrying out model training on the initial energy prediction model of the energy to be predicted by using each training data set to obtain each consumption prediction model of the energy to be predicted.

In the above embodiment, by constructing the training data set corresponding to the actual training requirement of each usage prediction model, the prediction accuracy of the usage prediction model obtained after the model training according to the training data set is effectively improved.

In one embodiment, the usage prediction model includes a first usage prediction model and a second usage prediction model; the step of constructing a training data set corresponding to each usage prediction model according to each historical characteristic data and the historical energy usage data, comprising: determining an initial training data set according to each historical characteristic data and the historical energy consumption data; based on a first data processing mode of the first quantity prediction model, carrying out data processing on the initial training data set to obtain a first training data set corresponding to the first quantity prediction model; and carrying out data processing on the initial training data set based on a second data processing mode of the second usage prediction model to obtain a second training data set corresponding to the second usage prediction model.

In the above embodiment, the data processing is performed on the initial training data set by the first data processing manner and the second data processing manner, so that a first training data set meeting the actual training requirement of the first usage prediction model and a second training data set meeting the actual training requirement of the second usage prediction model can be obtained, and a data basis is provided for effective training of the subsequent first usage training model and the second usage training model.

In one embodiment, the number of initial energy prediction models includes at least two; the training data sets are used for respectively carrying out model training on the initial energy prediction model of the energy to be predicted to obtain each consumption prediction model of the energy to be predicted, and the method comprises the following steps: respectively carrying out model training on at least two initial energy prediction models of the energy to be predicted based on different model training algorithms by using each training data set to obtain a plurality of candidate prediction models of the energy to be predicted; performing model prediction evaluation on each candidate prediction model according to the model evaluation parameters of the energy source to be predicted to obtain an evaluation result of each candidate prediction model; and determining each consumption prediction model of the energy source to be predicted from each candidate prediction model based on each evaluation result.

In the embodiment, the model training is performed on each initial energy prediction model by using different model training algorithms, and the usage prediction model with the optimal evaluation result is selected from the model training algorithms, so that the prediction accuracy and stability of the finally determined usage prediction model can be effectively improved, and the energy usage prediction accuracy and stability in actual usage prediction are further improved.

In a second aspect, the present application also provides an energy consumption prediction apparatus, the apparatus comprising: the production scheduling data acquisition module is used for acquiring characteristic production scheduling data corresponding to at least two consumption influence characteristics of the energy to be predicted; the model determining module is used for acquiring the energy consumption correlation of the consumption influence characteristics and the energy to be predicted according to each consumption influence characteristic; determining key impact features from each of the energy usage impact features based on each of the energy usage correlations; determining a target usage prediction model from at least two usage prediction models according to the key influence features and feature scheduling data of the key influence features in the feature scheduling data; the at least two consumption prediction models of the energy to be predicted are respectively used for representing different energy consumption trends of the energy to be predicted in the production process; and the consumption prediction module is used for predicting the consumption of the energy to be predicted according to the characteristic production data and the target consumption prediction model to obtain the predicted energy consumption of the energy to be predicted.

In a third aspect, the present application also provides a computer device comprising a memory storing a computer program and a processor implementing the steps of the method described above when the processor executes the computer program.

In a fourth aspect, the present application also provides a computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of the method described above.

In a fifth aspect, the application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method described above.

According to the energy consumption prediction method, the device, the computer equipment, the storage medium and the computer program product, when the energy consumption is predicted, the feature production data corresponding to each consumption influence feature influencing the energy consumption of the energy to be predicted can be obtained, the service production condition when the energy to be predicted is produced by using the energy to be predicted is reflected by the feature production data corresponding to each consumption influence feature, the target consumption prediction model matched with the feature production data is determined from at least two consumption prediction models obtained by training in advance, the target consumption prediction model can be more in accordance with the actual production condition when the energy to be predicted is produced by using the energy to be predicted, the consumption of the energy to be predicted is predicted according to each feature production data and the target consumption prediction model, more accurate predicted energy consumption can be obtained, and the accuracy of the energy consumption prediction of the energy to be predicted is effectively improved.

Drawings

FIG. 1 is a diagram of an application environment for a method of energy usage prediction in some embodiments;

FIG. 2 is a flow chart of a method of predicting energy usage in some embodiments;

FIG. 3 is a flow diagram of determining a target usage prediction model matching feature production data from at least two usage prediction models pre-trained based on feature production data in some embodiments;

FIG. 4 is a flow chart of determining a target predictor model from at least two predictors corresponding to key impact features according to the key impact features and feature scheduling data of the key impact features in some embodiments;

FIG. 5 is a flowchart of determining a target predictor model from at least two predictors corresponding to key impact features according to the key impact features and feature scheduling data of the key impact features in other embodiments;

FIG. 6 is a flow chart of a training method of each usage prediction model in some embodiments;

FIG. 7 is a flow chart of a training data set corresponding to each usage prediction model constructed according to each historical feature data and historical energy usage data in some embodiments;

FIG. 8 is a flow chart of model training an initial energy prediction model of an energy to be predicted using training data sets to obtain usage prediction models of the energy to be predicted;

FIG. 9 is a flow chart of a method for predicting energy usage in other embodiments;

FIG. 10 is a block diagram of an apparatus for predicting energy usage in some embodiments;

FIG. 11 is an internal block diagram of a computer device in some embodiments.

Detailed Description

Embodiments of the technical scheme of the present application will be described in detail below with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present application, and thus are merely examples, and are not intended to limit the scope of the present application.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having" and any variations thereof in the description of the application and the claims and the description of the drawings above are intended to cover a non-exclusive inclusion.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least some embodiments of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

In the description of the embodiments of the present application, the term "plurality" means two or more (including two), and similarly, "plural sets" means two or more (including two), and "plural sheets" means two or more (including two).

In the description of the embodiments of the present application, the orientation or positional relationship indicated by the technical terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. are based on the orientation or positional relationship shown in the drawings, and are merely for convenience of description and simplification of the description, and do not indicate or imply that the apparatus or element referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the embodiments of the present application.

In the description of the embodiments of the present application, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured" and the like should be construed broadly and may be, for example, fixedly connected, detachably connected, or integrally formed; or may be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communicated with the inside of two elements or the interaction relationship of the two elements. The specific meaning of the above terms in the embodiments of the present application will be understood by those of ordinary skill in the art according to specific circumstances.

In order to reduce the energy consumption generated by business production and improve the energy utilization rate, how to predict the energy consumption possibly generated during business production and manage the energy consumption based on the prediction result gradually becomes a core problem which needs to be faced by a plurality of enterprises in business production.

At present, prediction methods based on time series analysis or regression analysis and the like are generally adopted for predicting the energy consumption in the business production process, the accuracy of the methods is lower when the energy consumption is predicted for processing a large amount of complex data, and the prediction accuracy is lower when a small amount of production scheduling data is used for predicting the energy consumption, which may be caused by too little data in consideration of the consumption prediction.

In order to effectively improve the prediction accuracy of the energy consumption, when the energy consumption is predicted for the energy to be predicted, the feature production data corresponding to at least two consumption influence features of the energy to be predicted can be obtained, the feature production data corresponding to each of the plurality of consumption influence features can more accurately reflect the business production condition when the energy to be predicted is used for production, and the target consumption prediction model determined from at least two consumption prediction models obtained through pre-training is based on the feature production data corresponding to each of the consumption influence features, so that the target consumption prediction model can be more in accordance with the actual production condition when the energy to be predicted is used for production, and then the consumption of the energy to be predicted can be predicted according to each feature production data and the target consumption prediction model, so that the more accurate predicted energy consumption can be obtained, and the accuracy of the energy consumption prediction of the energy to be predicted is effectively improved.

The energy consumption prediction method provided by the embodiment of the application can be applied to an application environment shown in figure 1. Wherein usage prediction platform 102 communicates with business system 104 via a network. The data storage system may store data that the usage prediction platform 102 needs to process. The data storage system may be integrated on usage prediction platform 102 or may be located on the cloud or other network server. In the case that energy consumption prediction needs to be performed on the energy to be predicted, the consumption prediction platform 102 may acquire feature production data corresponding to at least two consumption influencing features of the energy to be predicted from the service system 104, determine, based on the feature production data, a target consumption prediction model matching with the feature production data from at least two consumption prediction models obtained by training in advance, and perform consumption prediction on the energy to be predicted according to the feature production data and the target consumption prediction model, to obtain the predicted energy consumption of the energy to be predicted.

The usage prediction platform 102 is a platform for performing usage prediction on the energy usage of at least one energy to be predicted according to production scheduling data in actual production, and the usage prediction platform 102 may be integrated on a terminal or a management system, where the terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The management system may be an energy management system used in the production of a service, for example in a power storage service scenario, the energy management system may be an EMS system. It will be appreciated that the management system may be implemented as a stand-alone server or as a cluster of servers.

The business system 104 is a management system for business production management, and can be used for realizing business production scheduling, production data storage, production monitoring, production management and control and other business management functions, and relevant production data for business production, such as scheduling data for current business production and historical production data generated in the historical business production process, are stored on the business system 104. The business system 104 may also be implemented as a stand-alone server or as a cluster of servers.

In some embodiments, as shown in fig. 2, an energy consumption prediction method is provided, and the method is applied to the consumption prediction platform in fig. 1 for illustration, and includes the following steps:

s202, acquiring characteristic production data corresponding to at least two consumption influence characteristics of the energy to be predicted.

The energy to be predicted refers to production and use energy, such as electric energy, natural gas, steam, water energy and the like, which need to be subjected to energy management by predicting the energy consumption during business production. The types and the amounts of the energy to be predicted, which are needed to be predicted, may be different according to different business production environments, for example, in a production environment using electric energy as a main consumption energy source, the energy to be predicted may be electric energy, and in a production environment using electric energy and water energy as main consumption energy sources, the energy to be predicted may be electric energy and water energy. It can be appreciated that when there are multiple energy sources to be predicted, a corresponding energy usage prediction operation may be performed for each energy source to be predicted.

The usage influencing characteristics of the energy to be predicted can be considered as usage influencing factors of the energy to be predicted, namely variable factors which can influence the usage of the energy to be predicted in production and the influence degree meets the condition of a preset degree, and when the characteristic data of the usage influencing characteristics change, the usage of the energy to be predicted also changes.

In actual production, there are various factors that may affect the amount of energy used, for example, in some embodiments, common influencing characteristics may include the yield of the product, the outdoor environment at the time of production, the indoor environment, the variety and system of the product, equipment opening variables, abnormal conditions, etc. Different energy sources to be predicted or the same energy source to be predicted may have different corresponding usage impact characteristics in different business production scenarios.

In some embodiments, the correspondence between the energy and the influence features is preset in the usage prediction platform, and after the usage prediction platform determines the energy to be predicted, the usage prediction platform may search for a plurality of usage influence features matching the energy to be predicted in the correspondence between the energy and the influence features.

In addition, in order to effectively improve the accuracy of energy consumption prediction of the energy to be predicted, the energy consumption influence characteristics of the energy to be predicted can be determined by influencing the correlation between the characteristics and the energy consumption of the energy to be predicted. For example, in some embodiments, the usage prediction platform may obtain historical feature data of each influencing feature, and historical energy usage data corresponding to the historical feature data, call a preset correlation calculation mode to determine a correlation between each influencing feature and an energy usage of the energy to be predicted based on the historical feature data and the historical energy usage data, and determine a plurality of usage influencing features of the energy to be predicted based on the correlation between each influencing feature and the energy usage of the energy to be predicted.

In some embodiments, the usage prediction platform may obtain a correlation sequence of each influence feature and the energy usage of the energy to be predicted in a descending order, and determine an influence feature corresponding to the correlation from the first position of the correlation sequence to the preset number of bits as the usage influence feature of the energy to be predicted. For example, in the case where the preset number of bits is 3, the influence feature corresponding to the correlation from the 1 st bit to the 3 rd bit in the correlation sequence may be determined as the usage influence feature of the energy to be predicted.

In some embodiments, the usage prediction platform may also compare the correlation between each influence feature and the energy usage of the energy to be predicted with a preset correlation threshold, and determine the influence feature corresponding to the correlation greater than or equal to the correlation threshold as the usage influence feature of the energy to be predicted. For example, in the case where the preset correlation threshold is 80%, among the correlations between the respective influence features and the energy usage of the energy to be predicted, the influence feature having the correlation of greater than or equal to 80% may be determined as the usage influence feature of the energy to be predicted.

The feature production data corresponding to the usage influencing features is data information which can reflect theoretical feature values of the usage influencing features in an actual production stage. The feature scheduling data corresponding to the usage influencing features can be determined by the service system when scheduling according to the actual production plan, so that the feature scheduling data corresponding to the usage features of the energy to be predicted can reflect the service production condition when the energy to be predicted is used for production.

In the field of business production, scheduling refers to a process of allocating production tasks to production resources, in some embodiments, after obtaining a production plan of business production, a business system may determine the production tasks according to the production plan, and then perform scheduling based on the production tasks and environmental conditions of a production environment, where usage influencing features will be allocated with corresponding feature scheduling data.

For example, when the volume-affecting feature is a volume, the business system may determine the volume of production that needs to be completed for the current production cycle based on the actual production plan, and determine the volume of production as feature production data for the volume-affecting feature. For another example, when the usage-affecting feature is an outdoor environment, the business system may determine a production cycle of the present round of production according to an actual production plan, determine outdoor environment data of the production environment in the production cycle according to the production cycle, and determine the outdoor environment data of the production environment as feature-producing data of the usage-affecting feature, which is the outdoor environment.

In some embodiments, in the case where there is an energy usage prediction demand for the energy to be predicted, the usage prediction platform may obtain feature production data corresponding to each of at least two usage-affecting features of the energy to be predicted from the service system.

In some embodiments, the usage prediction platform may determine at least two usage influencing features matched with the energy to be predicted from a preset correspondence between the energy to be predicted and the influencing features according to the energy to be predicted in response to an energy usage prediction instruction for the energy to be predicted, send a scheduling data acquisition instruction to the service system based on the usage influencing features, and receive feature scheduling data corresponding to the usage influencing features returned by the service system.

S204, determining a target usage prediction model matched with the feature production data from at least two usage prediction models obtained through pre-training based on the feature production data.

The consumption prediction model is a preset model for predicting the consumption of the energy to be predicted in the production process according to actual characteristic production data, and can represent the energy consumption trend of the energy to be predicted in the production process. The consumption prediction model can be obtained by training historical characteristic data corresponding to each consumption influence characteristic of the energy to be predicted and historical energy consumption data of the energy to be predicted in advance.

In the usage prediction platform, at least two usage prediction models are configured for each energy to be predicted, and it can be understood that model types of at least two usage prediction models of the same energy to be predicted are different, namely, at least two usage prediction models of the same energy to be predicted can respectively represent different energy consumption trends of the energy to be predicted in the production process. In some embodiments, the usage prediction models may be respectively corresponding to each business production situation using the energy to be predicted for production.

In some embodiments, when the usage prediction platform obtains the feature scheduling data, the usage prediction platform may determine, from at least two usage prediction models trained in advance for the energy to be predicted, a target usage prediction model matching with the feature scheduling data based on the business production situation during production reflected by the feature scheduling data, where the target usage prediction model may also be considered as a usage prediction model matching with the business production situation during subsequent actual production in each usage prediction model.

S206, according to the feature scheduling data and the target consumption prediction model, the consumption of the energy to be predicted is predicted, and the predicted energy consumption of the energy to be predicted is obtained.

In some embodiments, after determining the target usage prediction model of the energy to be predicted, the usage prediction platform may predict the usage of the energy to be predicted according to the obtained feature production data and the target usage prediction model, to obtain the predicted energy usage of the energy to be predicted.

In some embodiments, the usage prediction platform may input the feature scheduling data as an input parameter of the model into the target usage prediction model, and predict the usage of the energy to be predicted based on the target usage prediction model, so as to obtain the predicted energy usage of the energy to be predicted output by the target usage prediction model.

In some embodiments, in order to improve accuracy of usage prediction, the usage prediction platform may perform data preprocessing on each feature scheduling data based on a preset data preprocessing manner, process abnormal data, repeated data and the like that may occur in each feature scheduling data, input each feature scheduling data after processing as an input parameter of a model into a target usage prediction model, perform usage prediction on energy to be predicted based on the target usage prediction model, and obtain a predicted energy usage of the energy to be predicted output by the target usage prediction model.

In some embodiments, the usage prediction platform may further normalize the feature production data in order to reduce the risk of affecting the accuracy of the prediction due to a large magnitude of difference in the feature data corresponding to the usage-affecting features.

In the energy consumption prediction method, when the energy consumption prediction is carried out, the characteristic scheduling data corresponding to each consumption influence characteristic influencing the energy consumption of the energy to be predicted can be obtained, the service production condition when the energy to be predicted is produced by using the energy to be predicted reflected by the characteristic scheduling data corresponding to each consumption influence characteristic is determined from at least two consumption prediction models obtained by training in advance, the target consumption prediction model can be used for more conforming to the actual production condition when the energy to be predicted is produced by using the energy to be predicted, the consumption prediction of the energy to be predicted is carried out according to each characteristic scheduling data and the target consumption prediction model, more accurate predicted energy consumption can be obtained, and the accuracy of the energy consumption prediction of the energy to be predicted is effectively improved.

When the energy consumption is predicted, the accuracy of the energy consumption prediction is directly influenced by the selection of the target consumption prediction model, so that the determination of the target consumption prediction model is a key step for improving the prediction accuracy.

In some embodiments, S204, determining, based on each of the feature production data, a target usage prediction model that matches the feature production data from at least two usage prediction models trained in advance may include: the method comprises the steps of determining key influence features from all usage influence features, and determining a target usage prediction model from at least two usage prediction models based on the key influence features and feature production data of the key influence features in feature production data. The key influence features may be regarded as influence features having the highest priority among the usage influence features when the target usage prediction model is selected. Because a plurality of consumption influence features exist in the energy to be predicted, when model selection is performed based on feature scheduling data corresponding to the consumption influence features, the situation that different target consumption prediction models can be correspondingly selected by different consumption influence features may exist, so that a key consumption influence feature needs to be determined from the consumption influence features, and a unique target consumption prediction model is selected from the plurality of consumption prediction models through the key influence feature.

Further, in some of these embodiments, the key impact feature may be a model selection feature that the user determines based on actual production needs. For example, the user may determine the yield directly as a key impact feature, if the yield is determined to be an impact feature that is primarily considered in the overall business process. When the target consumption prediction model is determined, the consumption prediction platform can directly determine key influence characteristics of the energy to be predicted according to the characteristic identification of each consumption influence characteristic.

In other embodiments, the key impact feature may also be the usage impact feature that has the highest correlation with the energy source to be predicted. At this time, as shown in fig. 3, S204, based on each feature production data, determines a target usage prediction model matching the feature production data from at least two usage prediction models obtained by training in advance, including:

S302, according to each energy influence characteristic, energy consumption correlation between the energy influence characteristic and the energy to be predicted is obtained.

In some embodiments, the usage prediction platform may obtain, for each usage impact feature of the energy to be predicted, an energy usage correlation of each usage impact feature with the energy to be predicted, respectively.

In some embodiments, the energy consumption correlation between each energy consumption influence feature of the energy to be predicted and each energy consumption of the energy to be predicted is pre-stored in the consumption prediction platform, and the consumption prediction platform may directly obtain the energy consumption correlation between each energy consumption influence feature and the energy to be predicted from the storage area.

In some embodiments, a correlation calculation mode is preconfigured in the usage prediction platform, and the energy usage correlation between each usage influence feature and the energy to be predicted can be calculated in real time by using the correlation calculation mode according to the historical feature data of each usage influence feature and the historical energy usage data of the energy to be predicted.

S304, determining key influence features from the energy consumption influence features based on the energy consumption correlation.

In some embodiments, the usage prediction platform may determine a usage impact feature to which the maximum energy usage correlation belongs from among the energy usage correlations as a key impact feature of the energy to be predicted.

S306, determining a target usage prediction model from at least two usage prediction models according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data.

In some embodiments, after determining the key impact features of the energy to be predicted, the usage prediction platform may determine feature production data of the key impact features from the feature production data, and determine a target usage prediction model from the at least two usage prediction models based on the key impact features and the feature production data of the key impact features.

In the above embodiment, by determining the key influence features of the energy to be predicted and determining the target usage prediction model based on the key influence features, the number of the finally determined target usage prediction models can be unique, the determination efficiency of the target usage prediction model is effectively improved, and the prediction efficiency of energy usage prediction is further improved.

Since the usage prediction platform predicts based on a plurality of usage-affecting features of the energy source to be predicted when performing the usage prediction, in some embodiments, the usage prediction model includes a predictor model corresponding to each of the usage-affecting features, and model types of the predictor models corresponding to each of the usage-affecting features included in the usage prediction model are the same.

S306, determining a target usage prediction model from at least two usage prediction models according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data, wherein the method comprises the following steps:

And determining a target prediction sub-model from at least two prediction sub-models corresponding to the key influence features according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data. And determining the usage prediction model containing the target predictor model as a target usage prediction model.

The usage prediction model includes a predictor model corresponding to each usage influence feature of the energy to be predicted, and it can be understood that the predictor model can be used for representing the energy consumption trend of the energy to be predicted in the usage influence feature corresponding to the predictor model.

The model types of the predictor models belonging to one usage prediction model are the same as the model types of the predictor models belonging to one usage prediction model. The model types are the same, and it can be understood that the corresponding energy consumption trend trends are approximately the same, for example, if the energy consumption trend represented by a certain consumption prediction model is a steady ascending trend, then it can be considered that the energy consumption trend represented by all the predictor models in the consumption prediction model is also a steady ascending trend.

Because the energy source to be predicted corresponds to at least two consumption prediction models, and each consumption prediction model comprises a predictor model corresponding to each consumption influence characteristic, the key influence characteristic also corresponds to at least two predictor models, and each predictor model belongs to different consumption prediction models, namely, the model types of a plurality of predictor models corresponding to the same consumption influence characteristic are different.

In some embodiments, after determining the key impact feature and the feature scheduling data of the key impact feature, the usage prediction platform may determine a target predictor model from at least two predictors corresponding to the key impact feature based on the key impact feature and the feature scheduling data of the key impact feature, and determine a usage prediction model including the target predictor model as the target usage prediction model.

In this embodiment, at least two predictor models corresponding to the key influence feature are determined first, and then, the target predictor model is determined from the predictor models corresponding to the key influence feature directly according to the key influence feature and the feature production data of the key influence feature, so that the usage prediction model including the target predictor model can be determined as the target usage prediction model, thereby effectively improving the determination efficiency of the target usage prediction model and further improving the prediction efficiency of the energy usage prediction.

The energy consumption prediction model is used for the purpose of training a plurality of energy consumption prediction models in advance, so that the energy consumption prediction model of the energy to be predicted can cover various production conditions of the energy to be predicted in actual production, and the energy consumption prediction model with the energy consumption prediction vertex value is a prediction model with higher prediction precision in actual production.

In some embodiments, the usage prediction model includes a first usage prediction model including energy prediction vertex values, as shown in fig. 4, and determining a target prediction sub-model from at least two prediction sub-models corresponding to key impact features according to the key impact features and feature production data of the key impact features in the feature production data, including:

S402, determining model selection data of energy to be predicted according to energy prediction vertex values of predictor models corresponding to key influence features in the first quantity prediction model.

The first quantity prediction model with the energy prediction vertex value has higher prediction accuracy under certain production conditions, so that whether the first quantity prediction model can be used for quantity prediction can be prioritized when the target quantity prediction model is selected.

And the energy consumption trend represented by the first quantity prediction model is expressed in a curve form, so that the curve with the energy prediction vertex value comprises a trend ascending section and a trend descending section, different sections correspond to different production conditions, and the energy prediction vertex value can be considered as a boundary of different sections. It will be appreciated that, since the first quantity prediction model includes a plurality of predictor models, the first quantity prediction model also includes energy prediction vertex values of the plurality of predictor models.

In some embodiments, the usage prediction platform may determine a predictor model corresponding to the key impact feature from the first usage prediction model based on the key impact feature, and then determine model selection data for the energy to be predicted based on energy prediction vertex values of the predictor model corresponding to the key impact feature. Wherein the model selection data is model data for performing a use condition check of the first quantity prediction model. Model selection data may be determined based on energy prediction vertex values of the predictor models corresponding to the key impact features.

In some embodiments, the usage prediction platform may determine the historical feature data corresponding to the energy prediction vertex values as model selection data for the energy to be predicted.

In some embodiments, the usage prediction platform may determine, based on the energy prediction vertex value, a numerical interval formed by historical feature data in a trend rising interval in the first usage prediction model as model selection data of the energy to be predicted.

In some embodiments, the first energy prediction model may be a quadratic transformation model, and the energy consumption trend represented by the quadratic transformation model may be considered as a parabolic trend, so that there may be an energy consumption prediction vertex value, where the energy consumption prediction vertex value is located on a symmetry axis of a parabola.

S404, determining a predictor model corresponding to the key influence feature in the first quantity prediction model as a target predictor model when the feature production data of the key influence feature is determined to meet the use condition of the first quantity prediction model based on the model selection data.

The use condition of the first quantity prediction model is a preset judgment condition for judging whether the energy to be predicted can be predicted by using the first quantity prediction model.

In some embodiments, the usage prediction platform may determine, based on the model selection data, whether the feature production data of the key impact feature meets the usage condition of the first usage prediction model, and if it is determined that the feature production data of the key impact feature meets the usage condition of the first usage prediction model, indicate that the first usage prediction model matches the current feature production data, i.e., matches the current business generation situation. The usage prediction platform may determine a predictor model of the first usage prediction model that corresponds to the key impact feature as a target predictor model.

In some embodiments, when the model selection data is historical feature data corresponding to the energy prediction vertex value, the use condition of the first quantity prediction model may be that feature production data of the key influence feature is smaller than the model selection data, if the feature production data of the key influence feature meets the use condition of the first quantity prediction model, it may be explained that the feature production data of the key influence feature accords with an actual business production rule in an ascending trend, and a predictor model corresponding to the key influence feature in the first quantity prediction model may be determined as a target predictor model, that is, the first quantity prediction model is determined as a target quantity prediction model.

In some embodiments, in a case where the model selection data is a numerical interval constituted by historical feature data in a trend-up interval, the usage condition of the first quantity prediction model may be that feature production data of key influence features belongs to the numerical interval.

In the above embodiment, the model selection data of the energy to be predicted is determined by the energy prediction vertex value of the predictor model corresponding to the key influence feature, and whether the feature scheduling data of the key influence feature meets the use condition of the first quantity prediction model is firstly determined based on the model selection data, so that whether the current production condition of the energy to be predicted is matched with the first quantity prediction model can be quickly determined, the first quantity prediction model is directly used under the matched condition, and the accuracy of model determination can be provided while the model determination efficiency is improved.

In other embodiments, the usage prediction model further includes a second usage prediction model in which the energy usage prediction value is positively correlated with the feature production data of the usage affecting feature. As shown in fig. 5, the energy consumption prediction method further includes:

S502, determining a predictor model corresponding to the key influence feature in the second usage prediction model as a target predictor model when the feature production data of the key influence feature is determined to not meet the use condition of the first usage prediction model based on the model selection data.

The positive correlation between the energy predicted value and the feature production data of the usage influencing feature in the second usage prediction model means that the energy predicted value increases along with the increase of the feature production data of the usage influencing feature. From the graph, the energy consumption curve of the second usage prediction model can be considered as a continuously rising curve.

In some embodiments, the second usage prediction model may be a logarithmic transformation model, and the energy consumption trend represented by the logarithmic transformation model may be regarded as a linear upward trend, so as to solve the problem of the usage vertex existing in the first usage prediction model.

In some embodiments, when the usage prediction platform determines that the feature production data of the key impact feature does not meet the use condition of the first usage prediction model based on the model selection data, it is stated that the feature production data of the key impact feature falls into a trend-down interval of the first usage prediction model at this time, and as the feature production data increases, the energy prediction usage of the energy to be predicted decreases, which obviously does not conform to an actual production rule, so in order to make the usage prediction process of the energy to be predicted conform to the actual production rule, the usage prediction platform may determine, as a target prediction sub-model, a prediction sub-model corresponding to the key impact feature in the second usage prediction model, that is, a second usage prediction model in which the feature production data of the energy usage prediction value and the usage impact feature always have a positive correlation, as the target usage prediction model.

In this embodiment, under the condition that it is determined that the energy consumption prediction of the energy to be predicted by using the first consumption prediction model does not conform to the actual production rule, the energy consumption prediction is performed for the energy to be predicted by using the second consumption prediction model in which the energy consumption prediction value and the feature production data of the consumption influence feature are always in positive correlation, so that the energy consumption prediction process of the energy to be predicted conforms to the actual production rule more effectively, and the accuracy of the energy consumption prediction is improved.

In addition to normal production conditions, abnormal production conditions during production, such as special conditions, are also factors to be considered. Taking the first usage prediction model as a secondary transformation model and the second usage prediction model as a logarithmic transformation model as an example, in some embodiments, in the case that the business production is normal production, the energy consumption trend represented by the default model may be a parabolic trend with downward opening.

In other embodiments, in the case of abnormal production in the business production, for example, in the case of special working conditions such as tangential pulling, complex pulling, etc., the energy consumption trend represented by the model may be a parabolic trend with an upward opening, and in this case, the opening direction of the parabola needs to be considered when determining the target usage prediction model.

Thus, in some embodiments, it is desirable to first determine the direction of the opening of the parabola characterized by the predictor model for which the key impact feature corresponds. Under the condition that the parabola is provided with an upward opening, if the feature production data of the key influence features is smaller than the model selection data, the feature production data of the key influence features can be explained to be in a descending trend and do not accord with the actual business production rule, at the moment, the production can be considered to be in a special working condition, an abnormal working condition consumption prediction model corresponding to the special working condition can be called, and the energy consumption during the production can be accurately predicted.

And under the condition that the parabola is provided with an upward opening, if the characteristic production data of the key influence characteristic is larger than or equal to the symmetry axis of the parabola and smaller than the corresponding minimum yield when the front-stage working procedure and the rear-stage working procedure are fully opened, the secondary model can be directly adopted as a target consumption prediction model of the energy to be predicted. If the feature production data of the key influence features is larger than or equal to the corresponding minimum yield when the front-stage process and the back-stage process are fully opened, a logarithmic transformation model can be adopted as a target consumption prediction model of the energy to be predicted.

The usage influencing characteristics of the energy to be predicted may include influencing characteristics, such as equipment opening variables, which can directly obtain the characteristic production data, such as yield, outdoor environment, etc.

In some embodiments, in a case where the usage influencing feature includes an equipment start variable, a method for obtaining feature scheduling data corresponding to the equipment start variable includes:

Based on the device-related features satisfying the correlation condition with the device-on variable, related scheduling data of the device-related features is acquired. And processing the related scheduling data according to the energy to be predicted and the equipment prediction model matched with the equipment opening variable to obtain the characteristic scheduling data corresponding to the equipment opening variable.

The device opening variable refers to the number of devices to be opened in a production environment or the type and the number of the devices to be opened when actual production is carried out according to a production schedule.

The device-related feature refers to an influence feature that has a feature correlation with the usage-dependent feature of the device-on variable and that satisfies a correlation condition. It will be appreciated that the device correlation feature is an influence feature that may directly obtain feature production data, and may be one or more of a plurality of usage influence features of the energy source to be predicted, or may be other influence features, i.e. one or more of a plurality of usage influence features of the energy source not to be predicted.

Because the production is not actually started when the energy consumption is predicted, the equipment opening condition during the production is unknown, and the characteristic scheduling data of the equipment opening variable is required to be predicted according to the characteristic scheduling data of the equipment related characteristics with strong correlation with the equipment opening variable. For example, in some embodiments, yield may be considered as a device-related feature having a strong correlation with a device-on variable, and the feature-on-variable feature-on-data may be predicted from the yield feature-on-data.

The correlation condition is a preset judgment condition for judging whether the influence feature has strong correlation with the device on variable, and in some embodiments. The correlation condition may be that the correlation between the influencing feature and the device on variable is above a preset threshold, for example above 85%.

The device prediction model is used for predicting the device opening condition in actual production according to the feature scheduling data of the device related features, and can be obtained by training an initial prediction model in advance according to the historical feature data corresponding to the device related features and the historical variable data of the device opening variable. It can be understood that the device prediction models corresponding to the opening variables of the different energy devices to be predicted are different.

In some embodiments, when determining that the usage influencing feature of the energy source to be predicted includes the equipment start variable, the usage prediction platform may first determine equipment related features that satisfy the correlation condition with the equipment start variable, then acquire related scheduling data of the equipment related features, input the related scheduling data into an equipment prediction model matched with the equipment start variable, and process the related scheduling data through the equipment prediction model to acquire feature scheduling data output by the equipment prediction model, where the feature scheduling data sets feature scheduling data corresponding to the equipment start variable.

In some embodiments, the usage prediction platform sets in advance a device-related feature corresponding to a device-opening variable of each energy to be predicted, and may directly search for a device-related feature corresponding to the device-opening variable of the energy to be predicted from a preset correspondence between the device-opening variable and the device-related feature according to the energy identifier of the energy to be predicted.

In the above embodiment, for the equipment start variable that cannot directly obtain the feature scheduling data, the feature scheduling data of the equipment related features having strong correlation with the equipment start variable may be used, and the feature scheduling data of the equipment start variable is obtained by combining with the equipment prediction model obtained by pre-training, so that the consideration dimension of the usage influencing feature when the energy usage is predicted for the energy to be predicted can be effectively increased, and the accuracy of the energy usage prediction is further improved.

The improvement of the accuracy of the energy consumption prediction is not only dependent on the selection accuracy of the target consumption prediction model, but also the training effect of each consumption prediction model of the energy to be predicted is a great influence factor. The manner of training the usage prediction model will be described in several embodiments.

In some embodiments, as shown in fig. 6, the training mode of each usage prediction model includes the following steps:

s602, for each quantity influence characteristic of the energy to be predicted, historical characteristic data of each quantity influence characteristic and historical energy quantity data of the energy to be predicted are obtained.

The historical characteristic data of each usage influencing characteristic and the historical energy usage data of the energy to be predicted are actual production data generated and stored in the historical service production process of the service system.

In some embodiments, the usage prediction platform may obtain, for each usage impact feature of the energy to be predicted, historical feature data for each usage impact feature and historical energy usage data for the energy to be predicted from the business system when training a corresponding usage prediction model for the energy to be predicted.

In some embodiments, the usage prediction platform may obtain, from the service system, historical feature data of each usage influencing feature and historical energy usage data of the energy to be predicted according to a preset historical data period, and it may be understood that the historical data period may be determined by a designer according to an actual service production scenario, and by obtaining the historical feature data and the historical energy usage data in the historical data period, the data processing complexity may be reduced and the validity of the historical data may be improved to a certain extent.

In some embodiments, the usage prediction platform may also construct a corresponding historical database for each type of energy, for example, first determine candidate features that may have a correlation with the usage of each type of energy, then obtain historical feature data of each candidate feature and historical energy usage data of each type of energy from the service system according to the candidate features, construct a historical database corresponding to each type of energy, and then perform correlation analysis for each candidate feature and the energy usage of each type of energy, and determine each usage impact feature corresponding to each type of energy, thereby determining historical feature data and historical energy usage data corresponding to each type of energy.

In some embodiments, the usage prediction platform may use pearson correlation coefficients to calculate, for each type of energy, a correlation between each candidate feature and the energy usage, for historical feature data and historical energy usage for each candidate feature.

S604, constructing a training data set corresponding to each consumption prediction model according to each historical characteristic data and historical energy consumption data.

In some embodiments, because model types of the usage prediction models of each energy to be predicted are different, when model training is performed on each usage prediction model, data forms of training data used are also different, after each historical feature data and historical energy usage data are obtained, an initial training data set needs to be built by the usage prediction platform, then data transformation is performed on historical data in the initial training data set according to actual training requirements of each usage prediction model, and a corresponding training data set is built for each usage prediction model.

In some embodiments, because of possible production periodicity in the business production process, there may be a problem that the historical feature data and the historical energy consumption data obtained by the consumption prediction platform may not match, for example, when the production period is j days, the product produced by the consumed energy consumption in production needs to be completely formed on j days, and then the current energy consumption data should be actually matched with the production data after j days, so, in order to improve the matching between the historical feature data and the historical energy consumption data, the consumption prediction platform may determine, after obtaining each historical feature data and the historical energy consumption data, the historical energy consumption data substantially matched with each historical feature data according to the production period, for example, may match the extracted energy consumption data with the corresponding historical feature data after j days of rolling.

In some embodiments, in order to improve accuracy of model training, the usage prediction platform may further perform data preprocessing on the obtained historical energy usage data and the historical feature data, so as to improve data quality of training data, and further improve accuracy of model training.

The historical energy consumption data and the corresponding historical characteristic data are collectively referred to as historical data, for example, in some embodiments, the consumption prediction platform may perform missing value positioning on each historical data, determine missing values in each historical data, then determine data processing modes corresponding to each missing value according to data types of the missing values, and process each missing value based on the data processing mode corresponding to each missing value, so as to improve the integrity of the historical data. For example, for the missing value of the outdoor environment, the filling process may be performed by using historical data, and for the missing value of the yield, whether the yield is produced on the same day may be judged first, if the yield is produced on the same day, the collected yield data may be considered to have errors, and the accuracy of the yield data needs to be checked. If not produced the day, the missing value may be deleted directly.

In other embodiments, the usage prediction platform may further delete the repeated value and the abnormal value in the historical data, and take the abnormal value of the historical data as an example, the usage prediction platform may use, for example, a 3sigma method for the historical data, reject outliers in the historical data, and define the abnormal value according to the 3sigma principle when a certain sample is less than the mean minus 3 times the variance, or greater than the mean plus 3 times the variance, where the probability is 0.3%.

In some embodiments, for some historical data corresponding to abnormal working conditions, such as historical data of abnormal working conditions including cut-and-pull, redraw, debugging, specific production phase, energy consumption with yield, energy consumption without yield and the like, the consumption prediction platform can construct an independent database for the historical data of the abnormal working conditions, so that an abnormal working condition prediction model can be conveniently and independently built later, and an application energy standard corresponding to the abnormal working conditions is set.

S606, respectively carrying out model training on the initial energy prediction model of the energy to be predicted by using each training data set to obtain each consumption prediction model of the energy to be predicted.

The initial energy prediction model is an initial model which is not trained by a model, and has an initial prediction model framework.

In some embodiments, after each training data set corresponding to each usage prediction model is constructed, the usage prediction platform may use each training data set to perform model training on the initial energy prediction model of the energy to be predicted, so as to obtain each usage prediction model of the energy to be predicted. For example, if the energy to be predicted has 3 usage prediction models, the usage prediction platform needs to perform model training on the initial energy prediction models through 3 sets of training data sets, so as to obtain 3 usage prediction models after training.

In the above embodiment, by constructing the training data set corresponding to the actual training requirement of each usage prediction model, the prediction accuracy of the usage prediction model obtained after the model training according to the training data set is effectively improved.

As in actual use, in some embodiments, the usage prediction model may include a first usage prediction model having an energy prediction vertex value and a second usage prediction model having an energy usage prediction value that is positively correlated with the feature production data of the usage-affecting feature.

As shown in fig. 7, S604, constructing a training data set corresponding to each usage prediction model based on each historical feature data and historical energy usage data, includes:

S702, determining an initial training data set according to each historical characteristic data and historical energy consumption data.

In some embodiments, the usage prediction platform may determine an initial training data set from each of the historical feature data and the historical energy usage data, the initial training data set comprising a plurality of sets of matched historical feature data and historical energy usage data.

S704, performing data processing on the initial training data set based on a first data processing mode of the first quantity prediction model to obtain a first training data set corresponding to the first quantity prediction model.

The first data processing mode of the first quantity prediction model refers to a preset data processing mode capable of performing data processing on each training data in an initial training data set to obtain a training data set meeting actual training requirements of the first quantity prediction model.

In some embodiments, the usage prediction platform may determine a first data processing manner of the first usage prediction model according to the first usage prediction model, and then perform data processing on each initial training data in the initial training data set based on the first data processing manner, to obtain a first training data set corresponding to the first usage prediction model.

In some embodiments, the first quantitative prediction model may be a quadratic transformation model, and the corresponding first data processing manner may be performing a quadratic transformation on the initial training data.

S706, performing data processing on the initial training data set based on a second data processing mode of the second usage prediction model to obtain a second training data set corresponding to the second usage prediction model.

The second data processing mode of the second usage prediction model refers to a preset data processing mode capable of performing data processing on each training data in the initial training data set to obtain a training data set meeting the actual training requirement of the second usage prediction model.

In some embodiments, the usage prediction platform may determine a second data processing manner of the second usage prediction model according to the second usage prediction model, and then perform data processing on each initial training data in the initial training data set based on the second data processing manner, to obtain a second training data set corresponding to the second usage prediction model.

In some embodiments, the second usage prediction model may be a logarithmic transformation model, and the corresponding second data processing manner may be to perform logarithmic transformation processing on the initial training data.

In the above embodiment, the data processing is performed on the initial training data set by the first data processing manner and the second data processing manner, so that a first training data set meeting the actual training requirement of the first usage prediction model and a second training data set meeting the actual training requirement of the second usage prediction model can be obtained, and a data basis is provided for effective training of the subsequent first usage training model and the second usage training model.

In some embodiments, the number of initial energy prediction models includes at least two, each initial energy prediction model corresponding to a different model training algorithm. As shown in fig. 8, S606, using each training data set, model training is performed on an initial energy prediction model of an energy to be predicted, to obtain each usage prediction model of the energy to be predicted, including:

S802, performing model training on at least two initial energy prediction models of the energy to be predicted based on different model training algorithms by using each training data set to obtain a plurality of candidate prediction models of the energy to be predicted.

In order to improve the model prediction precision and stability of the usage prediction model, the usage prediction platform may be configured with a plurality of initial energy prediction models, and model training is performed on each initial energy prediction model through different algorithms in machine learning regression. Such as linear regression algorithms, polynomial regression algorithms, etc.

In some embodiments, the usage prediction platform may use each training data set to perform model training on at least two initial energy prediction models of the energy to be predicted based on different model training algorithms, respectively, to obtain a plurality of candidate prediction models of the energy to be predicted.

S804, carrying out model prediction evaluation on each candidate prediction model according to the energy model evaluation parameters to be predicted, and obtaining an evaluation result of each candidate prediction model.

The model evaluation parameters are index parameters for judging whether each candidate prediction model accords with the use requirement, the model evaluation parameters can be determined according to the actual use requirement of a user, for example, the user pursues higher model training precision, the model evaluation parameters can be set to be model training precision, for example, the user pursues higher stability, and the model evaluation parameters can be set to be model prediction stability. For another example, the model evaluation parameters can be set to the model training accuracy and stability at the same time.

In some embodiments, the usage prediction platform may perform model prediction evaluation on each candidate prediction model according to preset model evaluation parameters, to obtain an evaluation result of each candidate prediction model.

In one embodiment, taking model evaluation parameters including model training accuracy and stability as an example, the usage prediction platform may divide a training data set into a sample set, a test set and a verification set, input sample set data into each initial energy prediction model to perform model training, and an intermediate model obtained by training may predict the test set and the verification set. And calculating the deviation rate of the predicted value of the test set and the actual energy consumption value in the test set, and continuously using the verification set to verify the model stability of the energy prediction model under the condition that the deviation rate meets the use condition of the model, so as to obtain the model training precision and stability.

In some embodiments, the usage prediction platform needs to perform model tuning on the candidate prediction model under the condition that the deviation rate does not meet the model use condition, and performs model stability verification under the condition that the candidate prediction model is optimal, so as to obtain model training precision and stability.

In some embodiments, the method for calculating the deviation rate may be: deviation%= (target value/actual value-1) 100. The target value is an actual energy consumption value, and the actual value is a model predicted value.

In some of these embodiments, when the deviation rate is less than or equal to the preset threshold n, it may be determined that the deviation rate meets the model usage condition, and a model evaluation is available. When the deviation rate is larger than a preset threshold value n, it can be determined that the deviation rate does not meet the use condition of the model.

In some embodiments, the usage prediction platform may optimize model parameters of the candidate prediction model by using a grid search method, and when the regression forest is used, the grid search adjusts the tree of the tree, and the minimum She Zishu lamp parameters are adjusted to be optimal, so that the model is optimal. If the model parameters are simply adjusted and cannot meet the use conditions of the model, the consumption influence characteristics of the energy to be predicted can be readjusted, namely, the correlation threshold k is adjusted, the consumption influence characteristics of the energy to be predicted are redetermined, and the training data set is reconstructed based on the adjusted consumption influence characteristics for training. Finally, the model precision and stability can be improved by adding training sets.

S806, determining each consumption prediction model of the energy to be predicted from each candidate prediction model based on each evaluation result.

In some embodiments, the usage prediction model may determine each usage prediction model having the best evaluation result from among the candidate prediction models based on the evaluation results of each candidate prediction model. For example, each usage prediction model having optimal model accuracy and stability may be selected.

In the embodiment, the model training is performed on each initial energy prediction model by using different model training algorithms, and the usage prediction model with the optimal evaluation result is selected from the model training algorithms, so that the prediction accuracy and stability of the finally determined usage prediction model can be effectively improved, and the energy usage prediction accuracy and stability in actual usage prediction are further improved.

In some embodiments, as shown in fig. 9, an energy consumption prediction method is provided, and the method can be specifically divided into four parts, namely a data processing part, a feature engineering part, a model training part, and an output and presentation part.

In the data processing part, the method may specifically comprise the steps of:

s901, constructing a history database.

The consumption prediction platform can acquire historical feature data of each candidate feature and historical energy consumption data of each energy source to construct a historical database, wherein each candidate feature can comprise factory historical output, varieties and systems, outdoor environment and equipment opening quantity, and each historical energy consumption data can comprise historical data of electricity consumption, natural gas consumption, steam consumption and water consumption.

S902, performing data preprocessing on each history data in the history database, and updating each history data in the history database.

In the feature engineering section, the method may specifically include the steps of:

S903, performing correlation analysis on the candidate features and the energy consumption of each energy, and determining the candidate features with the correlation higher than a preset correlation threshold k as consumption influence features of each energy.

S904, rolling and matching the historical characteristic data corresponding to the usage influencing characteristics with the historical energy usage data of the energy, and respectively performing secondary transformation and log transformation on the matched historical data.

Wherein the number of rolling days for rolling matching can be determined according to the period of production products.

S905, performing standardization processing on each converted historical data to obtain secondary conversion historical data and log conversion historical data.

For example, the product order of magnitude is m bits, the outdoor environment order of magnitude is n bits, the equipment opening order of magnitude is 2 bits, and the orders of magnitude of each feature are greatly different, so that the sample data is subjected to standardized processing, and the influence of dimension on the consumption prediction model is reduced.

In the model training part, the method specifically comprises the following steps:

S906, dividing each historical data into a sample set, a test set and a verification set, and training to obtain a device prediction model according to the historical output data and the historical device starting data.

S907, training an initial energy prediction model through machine learning based on the sample set to obtain a candidate prediction model.

S908, calculating the deviation rate of each candidate prediction model as a criterion for evaluating whether the usage prediction model is available.

The method comprises the steps of inputting a test set into each candidate prediction model to conduct prediction, obtaining a test set prediction value, calculating the deviation rate of each candidate prediction model according to the test set prediction value and the actual energy consumption value of the test set, and conducting model stability verification on each candidate prediction model by using a verification set.

S909, judging whether the deviation rate exceeds a preset threshold value n.

And S910, if the candidate prediction model is exceeded, tuning and optimizing the candidate prediction model.

The dosage prediction platform can optimize the optimal parameters of the algorithm model machine through the super parameters of the grid search adjustment method. The change of characteristics can also be re-made to the variables of the history data. The model performance can also be improved by adding training sets.

If not, S911 may be performed for subsequent outputting and rendering of the portion.

In the output and presentation part, the method may specifically include the steps of:

S912, obtaining feature scheduling data of each usage influencing feature.

And the consumption prediction platform calls an equipment prediction model according to the characteristic scheduling data of the yield, and predicts to obtain future equipment starting data.

S913, determining a target consumption prediction model from the consumption prediction models according to the relation that the larger the yield is, the larger the energy consumption is.

And determining the yield as a key influence characteristic, wherein the larger the yield is, the larger the energy consumption is, so that a consumption prediction model corresponding to the yield is determined from the secondary transformation model according to the yield, a parabolic symmetry axis corresponding to the yield is determined based on the consumption prediction model corresponding to the yield, and then the relation between the yield and the parabolic symmetry axis of future yield is determined according to the characteristic yield data of the yield.

As shown in fig. 10, in the case where the yield of future production is smaller than the parabolic symmetry axis, a quadratic transformation model, that is, a usage prediction model of each extracted feature quadratic transformation is directly adopted as a target usage prediction model of the energy to be predicted. In the case that the yield of future production is greater than or equal to the parabolic symmetry axis, the logarithmic transformation model, that is, the usage prediction model of each extracted feature log transformation, may be used as the target usage prediction model of the energy to be predicted.

When the key impact features are other features, such as an outdoor environment, the business logic of the outdoor environment and the energy consumption and the decision logic of the output and the energy consumption are consistent, and are not described herein.

S914, inputting the feature scheduling data corresponding to all the consumption influencing features into a target consumption prediction model to predict the energy consumption under different conditions.

S915, embedding and deploying each usage prediction model obtained through training in the energy management system for testing by a user.

The whole codes of the consumption prediction models can be deployed in an energy management system, and a user can select the energy type and the variable suggested energy consumption model in a self-defined manner and then test the energy type and the variable suggested energy consumption model.

Finally, after the usage prediction models are deployed in the energy management system, the application of the energy usage prediction method can be performed, for example, the predicted energy usage obtained by prediction is used as a management target, an alarm prompt is sent out under the condition that the actual energy usage exceeds the management target, the root cause exceeding the management target is deeply dug according to the alarm prompt, a knowledge base is formed according to various root causes possibly existing when the actual energy usage exceeds the management target, the knowledge base is triggered in future production, effective measures can be taken to reduce energy consumption, improve energy utilization efficiency, and achieve the purposes of saving energy, reducing consumption and reducing cost. For example, the production data in actual production can be adjusted and optimized according to the predicted energy consumption obtained by prediction and the energy consumption limit in the production process, so that the energy consumed in actual production can meet the energy consumption limit requirement.

In actual production, the energy consumption prediction is beneficial to a producer to better grasp the energy condition, reasonably distribute and use energy, so that the economic benefit is improved, the energy demand and the proportion relation thereof can be more accurately known, the energy structure and the layout are adjusted in a targeted manner, the reasonable allocation of resources is realized, the cost accounting and the department accounting are beneficial to the producer, the production is better organized, and the production efficiency is improved. Through the real-time monitoring, analysis and control of the energy use, the producer can realize the saving and reasonable utilization of the energy and reduce the production cost. In this embodiment, by combining the above parts, the spatial and temporal features in the data can be effectively utilized, and the prediction accuracy and stability can be improved. The energy consumption prediction model is trained in a regression learning mode, features in data can be automatically extracted and predicted, and manual intervention and calculation cost are reduced. In addition, the energy consumption prediction model has good expansibility, and can be suitable for the energy consumption prediction problem of different fields and scenes.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides an energy consumption prediction device for realizing the above related energy consumption prediction method. The implementation of the solution provided by the device is similar to that described in the above method, so the specific limitation of the embodiment of the energy consumption prediction device or devices provided below may be referred to as the limitation of the energy consumption prediction method hereinabove, and will not be repeated herein.

In some embodiments, as shown in fig. 10, there is provided an energy usage prediction apparatus 1000 comprising: a production data acquisition module 1001, a model determination module 1002, and a usage prediction module 1003, wherein:

the production data obtaining module 1001 is configured to obtain feature production data corresponding to at least two usage affecting features of the energy to be predicted.

The model determining module 1002 is configured to determine, based on the feature production data, a target usage prediction model that matches the feature production data from at least two usage prediction models that are obtained through training in advance.

And the consumption prediction module 1003 is configured to predict the consumption of the energy to be predicted according to the feature production data and the target consumption prediction model, so as to obtain the predicted energy consumption of the energy to be predicted.

In some embodiments, the model determination module is to: aiming at each quantity influence characteristic, acquiring the energy quantity correlation between the quantity influence characteristic and the energy to be predicted; determining key influence features from the usage influence features based on the energy usage correlations; and determining a target usage prediction model from the at least two usage prediction models according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data.

In some embodiments, the usage prediction model includes a predictor model that corresponds to each usage impact feature, respectively, and the model type of the predictor model that corresponds to each usage impact feature included in the usage prediction model is the same. The model determination module is used for: and determining a target prediction sub-model from at least two prediction sub-models corresponding to the key influence features according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data. And determining the usage prediction model containing the target predictor model as a target usage prediction model.

In some embodiments, the usage prediction model includes a first usage prediction model in which energy prediction vertex values exist. The model determination module is used for: determining model selection data of energy to be predicted according to energy prediction vertex values of a predictor model corresponding to key influence features in the first quantity prediction model; and determining a predictor model corresponding to the key influence feature in the first quantity prediction model as a target predictor model under the condition that the feature production data of the key influence feature is determined to meet the use condition of the first quantity prediction model based on the model selection data.

In some embodiments, the usage prediction model further includes a second usage prediction model in which the energy usage prediction value is positively correlated with the feature production data of the usage affecting feature. The model determination module is further to: and determining a predictor model corresponding to the key influence feature in the second usage prediction model as a target predictor model under the condition that the feature production data of the key influence feature is determined to not meet the use condition of the first usage prediction model based on the model selection data.

In some embodiments, where the usage influencing feature comprises a device turn-on variable, the production data acquisition module is further to: based on the device-related features satisfying the correlation condition with the device-on variable, related scheduling data of the device-related features is acquired. And processing the related scheduling data according to the energy to be predicted and the equipment prediction model matched with the equipment opening variable to obtain the characteristic scheduling data corresponding to the equipment opening variable.

In some embodiments, the energy usage prediction device further comprises:

The model training module is used for acquiring historical characteristic data of each consumption influence characteristic and historical energy consumption data of the energy to be predicted according to each consumption influence characteristic of the energy to be predicted; according to the historical characteristic data and the historical energy consumption data, constructing a training data set corresponding to each consumption prediction model; and respectively carrying out model training on the initial energy prediction model of the energy to be predicted by using each training data set to obtain each consumption prediction model of the energy to be predicted.

In some embodiments, the usage prediction model includes a first usage prediction model and a second usage prediction model. The model training module is used for: determining an initial training data set according to the historical characteristic data and the historical energy consumption data; based on a first data processing mode of the first quantity prediction model, performing data processing on the initial training data set to obtain a first training data set corresponding to the first quantity prediction model; and carrying out data processing on the initial training data set based on a second data processing mode of the second usage prediction model to obtain a second training data set corresponding to the second usage prediction model.

In some embodiments, the number of initial energy prediction models includes at least two. The model training module is used for: respectively carrying out model training on at least two initial energy prediction models of the energy to be predicted by using each training data set to obtain a plurality of candidate prediction models of the energy to be predicted; carrying out model prediction evaluation on each candidate prediction model according to the energy model evaluation parameters to be predicted to obtain an evaluation result of each candidate prediction model; and determining each consumption prediction model of the energy to be predicted from each candidate prediction model based on each evaluation result.

The modules in the energy consumption prediction device can be realized in whole or in part by software, hardware and a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In some embodiments, a computer device is provided, which may be a server integrated with a usage prediction platform, the internal structure of which may be as shown in FIG. 11. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer equipment is used for storing data such as consumption influencing characteristics, characteristic scheduling data, consumption prediction models, target consumption prediction models, predicted energy consumption and the like. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of energy usage prediction.

It will be appreciated by those skilled in the art that the structure shown in FIG. 11 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In some embodiments, a computer device is provided, comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing specific steps of an embodiment of the energy usage prediction method described above when the computer program is executed.

In some embodiments, a computer readable storage medium is provided, on which a computer program is stored which, when executed by a processor, implements the specific steps of the energy usage prediction method embodiments described above.

In some embodiments, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the specific steps of the energy usage prediction method embodiments described above.

The user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party. And, the acquisition, storage, processing, transmission and the like of the data all accord with the relevant regulations of laws and regulations.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magneto-resistive random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (PHASE CHANGE Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in various forms such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims (11)

1. A method of energy usage prediction, the method comprising:

Acquiring characteristic production data corresponding to at least two consumption influence characteristics of the energy to be predicted;

For each consumption influence characteristic, acquiring the energy consumption correlation of the consumption influence characteristic and the energy to be predicted;

determining key impact features from each of the energy usage impact features based on each of the energy usage correlations; the key influence features are consumption influence features with highest correlation with the energy to be predicted; determining a target usage prediction model from at least two usage prediction models according to the key influence features and feature scheduling data of the key influence features in the feature scheduling data; the at least two consumption prediction models of the energy to be predicted are respectively used for representing different energy consumption trends of the energy to be predicted in the production process;

According to the characteristic scheduling data and the target consumption prediction model, predicting the consumption of the energy to be predicted to obtain the predicted energy consumption of the energy to be predicted;

The usage prediction model comprises predictor models respectively corresponding to the usage influence features, and model types of the predictor models respectively corresponding to the usage influence features included in the usage prediction model are the same;

The consumption prediction model comprises a first consumption prediction model with an energy prediction vertex value and a second consumption prediction model, wherein the energy consumption prediction value in the second consumption prediction model is positively correlated with the characteristic production data of the consumption influence characteristic; the determining the target usage prediction model from at least two usage prediction models according to the key influence feature and the feature production data of the key influence feature in the feature production data comprises:

Determining a target prediction sub-model from at least two prediction sub-models corresponding to the key influence features according to the key influence features and the feature scheduling data of the key influence features in the feature scheduling data;

And determining a usage prediction model containing the target predictor model as the target usage prediction model.

2. The method of claim 1, wherein the step of determining the position of the substrate comprises,

The determining a target prediction sub-model according to the key influence feature and the feature scheduling data of the key influence feature in the feature scheduling data from at least two prediction sub-models corresponding to the key influence feature includes:

determining model selection data of the energy to be predicted according to an energy prediction vertex value of a predictor model corresponding to the key influence feature in the first quantity prediction model;

And determining a predictor model corresponding to the key influence feature in the first quantity prediction model as the target predictor model under the condition that the feature production data of the key influence feature meets the use condition of the first quantity prediction model based on the model selection data.

3. The method of claim 2, wherein the step of determining the position of the substrate comprises,

The method further comprises the steps of:

And determining a predictor model corresponding to the key influence feature in the second usage prediction model as a target predictor model under the condition that the feature production data of the key influence feature is determined not to meet the usage condition of the first usage prediction model based on the model selection data.

4. A method according to any one of claims 1-3, wherein, in case the usage influencing feature comprises a device on variable, the means for obtaining feature production data corresponding to the device on variable comprises:

acquiring related scheduling data of the device related features based on the device related features meeting the correlation conditions with the device opening variable;

and processing the related production scheduling data according to the energy to be predicted and the equipment prediction model matched with the equipment opening variable to obtain the characteristic production scheduling data corresponding to the equipment opening variable.

5. The method of claim 1, wherein the training of each of the usage prediction models comprises:

for each consumption influence characteristic of the energy to be predicted, acquiring historical characteristic data of each consumption influence characteristic and historical energy consumption data of the energy to be predicted;

according to the historical characteristic data and the historical energy consumption data, constructing a training data set corresponding to each consumption prediction model;

And respectively carrying out model training on the initial energy prediction model of the energy to be predicted by using each training data set to obtain each consumption prediction model of the energy to be predicted.

6. The method of claim 5, wherein the usage prediction model comprises a first usage prediction model and a second usage prediction model;

The step of constructing a training data set corresponding to each usage prediction model according to each historical characteristic data and the historical energy usage data, comprising:

Determining an initial training data set according to each historical characteristic data and the historical energy consumption data;

based on a first data processing mode of the first quantity prediction model, carrying out data processing on the initial training data set to obtain a first training data set corresponding to the first quantity prediction model;

and carrying out data processing on the initial training data set based on a second data processing mode of the second usage prediction model to obtain a second training data set corresponding to the second usage prediction model.

7. The method of claim 5 or 6, wherein the number of initial energy prediction models comprises at least two;

The training data sets are used for respectively carrying out model training on the initial energy prediction model of the energy to be predicted to obtain each consumption prediction model of the energy to be predicted, and the method comprises the following steps:

Respectively carrying out model training on at least two initial energy prediction models of the energy to be predicted based on different model training algorithms by using each training data set to obtain a plurality of candidate prediction models of the energy to be predicted;

performing model prediction evaluation on each candidate prediction model according to the model evaluation parameters of the energy source to be predicted to obtain an evaluation result of each candidate prediction model;

And determining each consumption prediction model of the energy source to be predicted from each candidate prediction model based on each evaluation result.

8. An energy usage prediction device, the device comprising:

The production scheduling data acquisition module is used for acquiring characteristic production scheduling data corresponding to at least two consumption influence characteristics of the energy to be predicted;

the model determining module is used for acquiring the energy consumption correlation of the consumption influence characteristics and the energy to be predicted according to each consumption influence characteristic; determining key impact features from each of the energy usage impact features based on each of the energy usage correlations; the key influence features are consumption influence features with highest correlation with the energy to be predicted; determining a target usage prediction model from at least two usage prediction models according to the key influence features and feature scheduling data of the key influence features in the feature scheduling data; the at least two consumption prediction models of the energy to be predicted are respectively used for representing different energy consumption trends of the energy to be predicted in the production process; the usage prediction model comprises predictor models respectively corresponding to the usage influence features, and model types of the predictor models respectively corresponding to the usage influence features included in the usage prediction model are the same; the consumption prediction model comprises a first consumption prediction model with an energy prediction vertex value and a second consumption prediction model, wherein the energy consumption prediction value in the second consumption prediction model is positively correlated with the characteristic production data of the consumption influence characteristic; the model determining module is further used for determining a target predictor model from at least two predictor models corresponding to the key influence features according to the key influence features and feature production data of the key influence features in the feature production data; determining a usage prediction model comprising the target predictor model as the target usage prediction model;

and the consumption prediction module is used for predicting the consumption of the energy to be predicted according to the characteristic production data and the target consumption prediction model to obtain the predicted energy consumption of the energy to be predicted.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.