CN115358441A - New energy cluster consumption intelligent control method and device based on federated learning - Google Patents

Abstract

The invention provides a new energy cluster consumption intelligent regulation and control method and a device based on federal learning, wherein the method comprises the following steps: collecting data of different new energy power generation equipment and power users, and establishing a multi-time scale index set, wherein the multi-time scale index set comprises power generation and power utilization characteristic index sets with different time scales; based on the multi-time scale index set, evaluating the potential of the adjustable resource participating in response scheduling to obtain a potential evaluation result of the adjustable resource; and aiming at the potential evaluation result of the adjustable resources, intelligently scheduling the resources of the whole network based on federal learning. According to the method, under the influence of uncertain factors, the cluster regulation potential evaluation of the new energy equipment and the user is realized, the data privacy of each area in the subsequent resource regulation process is protected based on the federal learning thought, and the safety of the overall intelligent regulation is improved.

Description

New energy cluster consumption intelligent control method and device based on federated learning

technical field

The invention relates to the field of resource regulation and control, in particular to a method and device for intelligent regulation and control of new energy cluster consumption based on federated learning.

Background technique

Uncertain factors that affect the response potential of adjustable resources include temperature, meteorological environment, operating status of adjustable resources, response behavior, and response incentive price. These uncertain factors reduce the reliability of the adjustment output of the adjustable resources participating in the power grid balance business, and intensify the fluctuation of the adjustment output of the adjustable resource response. Therefore, a large number of uncertain factors in the adjustable resource cluster limit the accuracy of the response potential assessment process, and it is very important to properly describe and deal with various uncertainties for the response potential assessment. The current research is mainly based on various types of adjustable resource aggregation response models to consider the impact of the market, the environment, and the power consumption characteristics of adjustable resources on the response potential, and consider the fluctuation of multi-cluster coordinated response output of adjustable resources. However, the research on the response potential of adjustable resource clusters mainly focuses on the research on the response capacity of adjustable resources, without considering the dynamic process of internal resources in time when the cluster participates in the response, and has no effect on the dynamic performance changes of adjustable resource clusters under complementary and coordinated operation. further analysis. Therefore, it is necessary to consider the dynamic process of cluster participation in the response, and evaluate the response potential of the complementary and coordinated operation of large-scale regional adjustable resource clusters from different time scales.

When regulating source-load resources based on potential evaluation results, different regions have their own data, which need to be used for training updates of actual models and optimization regulation calculations. However, the amount of data in a single region is insufficient, and the effect of model training is rough. If data from different regions are integrated for overall model training, it will be limited by the privacy protection of user data in practice. In principle, each regional center cannot leak user data to third parties at will. Distributed computing can be used to solve this problem. Federated learning is a new computing idea developed by distributed learning. Compared with distributed computing, the advantages of federated learning are mainly reflected in the following five aspects:

1. In terms of control degree, in traditional distributed learning, the central server has absolute control over the workers in each sub-region; in federated learning, users have absolute control over equipment and data, and can stop participating in computing at any time.

2. In terms of node stability, distributed worker nodes are very stable, with almost the same performance, which is too idealized; worker nodes in federated learning are unstable and different, such as mobile phones with different network speeds or different devices, computing speed performance It is different and more suitable for actual equipment and user situations.

3. In terms of data nature, the data of each node in traditional distributed distribution is similar, and the calculation after random scrambling has no major impact; the data of federated learning is not independent and identically distributed, and the nature of each node's data is different, taking into account the different user habits. A practical issue.

4. In terms of node data load, traditional distributed computing needs to ensure load balance; the data load of federated nodes can be unbalanced, and different units of electric energy have different weights, and the calculation speed can be flexibly adjusted according to node data.

5. In terms of regulation costs, distributed computing is mainly the computing cost of each worker node; federated learning costs are mainly concentrated in the communication cost between worker nodes and server nodes.

Contents of the invention

Aiming at the above-mentioned deficiencies in the prior art, the present invention provides a new energy cluster consumption intelligent control method and device based on federated learning, which realizes the cluster control potential evaluation of new energy equipment and users under the influence of uncertain factors , and based on the idea of federated learning, it protects the data privacy of each region in the subsequent resource regulation process and improves the security of the overall intelligent regulation.

A federated learning-based intelligent control method for new energy cluster consumption, comprising the following steps:

Collect data of different new energy power generation equipment and power users, and establish a multi-time scale index set, the multi-time scale index set includes power generation and power consumption characteristic index sets of different time scales;

Based on the multi-time scale index set, evaluate the potential of the adjustable resource to participate in the response scheduling, and obtain the potential evaluation result of the adjustable resource;

According to the potential evaluation results of the adjustable resources, the resources of the whole network are intelligently scheduled based on federated learning.

Further, collect data of different new energy power generation equipment and power users, consider the uncertainty of source-load resource regulation, and establish a multi-time scale index set, the multi-time scale index set includes power generation and power consumption characteristic index sets of different time scales , including the following steps:

Determine the external factors affecting the dispatch potential uncertainty of new energy output resources, collect daily data, and establish a set of power generation characteristic indicators on different time scales;

The external influencing factors of the dispatch potential uncertainty include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunshine and snow. The feature index set is obtained through statistical calculation of daily power generation data; the establishment module of the electricity consumption characteristic index set is used to determine the internal factors affecting the dispatch potential uncertainty of user demand response resources, collect data, and establish electricity consumption characteristics at different time scales indicator set;

The internal influencing factors of the scheduling potential uncertainty include the equivalent heat capacity C of the polymer, the equivalent thermal resistance R, the energy efficiency ratio η, the set temperature T _set , the allowable temperature adjustment ΔT, the control duration Δt, and the user's water consumption m _n , ambient temperature T _out , market electricity price TOU, and the power consumption characteristic index sets at different time scales are obtained through statistical calculation of daily power generation data.

Further, based on the multi-time scale index set, considering the uncertainty of source-load resources, evaluate the potential of adjustable resources participating in response scheduling, and obtain the potential evaluation results of adjustable resources, specifically including the following steps:

Establish an assessment model for adjustable resource response potential:

标准差为

参数μ₀与σ₀为根据r₄的历史响应数据集进行点估计得到的概率估计值；Among them, r ₁ , r ₂ , r ₃ are known deterministic model parameters; r ₄ is a normal distribution that satisfies certain rules, and the mean of the distribution is

standard deviation is

Parameters μ ₀ and σ ₀ are probability estimates obtained by point estimation based on the historical response data set of r ₄ ;

The obtained probability estimate is regarded as a normal distribution parameter satisfied by _r4 , so as to obtain the potential evaluation result of the adjustable resource participating in the response scheduling:

Among them, the mean reference value μ, the variance reference value σ ² , the mean fuzzy value μ _e and the variance fuzzy value σ _e are respectively model indicators describing the uncertainty parameters of user response potential. Uncertainty about the potential for business adjustment.

Further, according to the potential evaluation results of the adjustable resources, intelligent scheduling of the resources of the entire network is performed based on federated learning, specifically including:

With the goal of optimizing the overall effect of new energy consumption, a particle swarm optimization model is established:

The specific optimization objective is shown in formula (6), including reducing power fluctuations and reducing new energy curtailment. The constraints are shown in formula (8), including equipment output limitations and power fluctuation rate limitations.

Among them, T _new represents the number of periods for calculating power fluctuation, λ _o,t is the purchase price of electric energy, λ _flu is the cost coefficient of power fluctuation loss, _PG,t is the actual output of new energy, and P _Go,t is the output capacity of new energy , _PGmax,t and _PGmin,t represent the upper and lower bounds of new energy equipment output in period t, δ _t represents the fluctuation of new energy output power in period t, _δrise is the upper limit of power increase rate, and _δdrop is power reduction The lower limit of the rate, Δt is the time interval between two periods.

Furthermore, the server of the general control center uses the idea of federated learning to control the resources of the entire network, including:

Step (a) the general control center server transmits the parameter w of the fixed particle swarm optimization algorithm model to each sub-regional center;

Step (b) Based on the potential evaluation results of the adjustable resources, the workers in each sub-regional center obtain the power data of the local adjustable resources participating in demand response, and use the source-load power data information and the model parameters transmitted from the general control center w, sub-regional center workers perform local conventional particle swarm optimization respectively, and calculate the new energy output of each region or the adjustment amount of user load energy consumption;

Step (c) Each sub-regional center worker uses local data to perform multiple gradient descents locally, obtains the updated gradient g, and calculates the new parameter w and returns it to the general control center server.

w←w-α·g (8)

Among them, w is the particle swarm optimization model parameter, α is the specific weight coefficient update step size, and g is the gradient;

Step (d) The general control center server collects the new parameters w returned by the workers in each area, performs weighted average on the parameters, updates the parameters of the particle swarm model, and then sends them to the workers in each sub-area center;

Step (e) Repeat steps (b)-(d) until the controllable source-load resource scheduling in each area is completed.

An intelligent control device for new energy cluster consumption based on federated learning, including:

A characteristic index set establishment module, which is used to collect data of different new energy power generation equipment and power users, and establish a multi-time scale index set, the multi-time scale index set includes power generation and power consumption characteristic index sets of different time scales;

A potential evaluation module, configured to evaluate the potential of adjustable resources to participate in response scheduling based on the multi-time scale index set, and obtain the potential evaluation results of adjustable resources;

The intelligent scheduling module is used to intelligently schedule the resources of the entire network based on federated learning based on the potential evaluation results of the adjustable resources.

The characteristic index set establishment module includes a power generation characteristic index set establishment module and a power consumption characteristic index set establishment module;

The power generation characteristic index set building module is used to determine the external influence factors of the uncertainty of dispatch potential of new energy output resources, collect daily data, and establish power generation characteristic index sets of different time scales;

The external influencing factors of the dispatch potential uncertainty include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunshine and snow. The feature index set is obtained through statistical calculation of daily power generation data;

The establishment module of the electricity consumption characteristic index set is used to determine the internal influencing factors of the dispatch potential uncertainty of user demand response resources, collect data, and establish electricity consumption characteristic index sets of different time scales.

The internal influencing factors of the scheduling potential uncertainty include the equivalent heat capacity C of the polymer, the equivalent thermal resistance R, the energy efficiency ratio η, the set temperature T _set , the allowable temperature adjustment ΔT, the control duration Δt, and the user's water consumption m _n , ambient temperature T _out , market electricity price TOU, and the power consumption characteristic index sets at different time scales are obtained through statistical calculation of daily power generation data.

Further, the potential assessment module includes a potential assessment model building module and a potential assessment result acquisition module; wherein,

The potential assessment model building module is used to establish an adjustable resource response potential assessment model:

标准差为

参数μ₀与σ₀为根据r₄的历史响应数据集进行点估计得到的概率估计值；Among them, r ₁ , r ₂ , r ₃ are known deterministic model parameters; r ₄ is a normal distribution that satisfies certain rules, and the mean of the distribution is

standard deviation is

Parameters μ ₀ and σ ₀ are probability estimates obtained by point estimation based on the historical response data set of r ₄ ;

The potential evaluation result acquisition module is used to treat the obtained probability estimate as a normal distribution parameter satisfied by r ₄ , so as to obtain the potential evaluation result of the adjustable resource participating in the response scheduling:

Among them, the mean reference value μ, the variance reference value σ ² , the mean fuzzy value μ _e and the variance fuzzy value σ _e are respectively model indicators describing the uncertainty parameters of user response potential. Uncertainty about the potential for business adjustment.

Further, the intelligent scheduling module includes a particle swarm optimization model establishment module and a resource regulation module of the whole network;

The particle swarm optimization model establishment module is used to establish a particle swarm optimization model with the goal of optimizing the overall effect of new energy consumption.

Further, the resource control module of the whole network includes the general control center server and the regional center worker;

The general control center server is used to use the idea of federated learning to control the resources of the whole network, including:

Step (a) the general control center server transmits the parameter w of the fixed particle swarm optimization algorithm model to each sub-regional center;

Step (b) Based on the potential evaluation results of the adjustable resources, the workers in each sub-regional center obtain the power data of the local adjustable resources participating in demand response, and use the source-load power data information and the model parameters transmitted from the general control center w, sub-regional center workers perform local conventional particle swarm optimization respectively, and calculate the new energy output of each region or the adjustment amount of user load energy consumption;

Step (c) Each sub-regional center worker uses local data to perform multiple gradient descents locally, obtains the updated gradient g, and calculates the new parameter w and returns it to the general control center server.

w←w-α·g (9)

Among them, w is the particle swarm optimization model parameter, α is the specific weight coefficient update step size, and g is the gradient;

Step (d) The general control center server collects the new parameters w returned by the workers in each area, performs weighted average on the parameters, updates the parameters of the particle swarm model, and then sends them to the workers in each sub-area center;

Step (e) Repeat steps (b)-(d) until the controllable source-load resource scheduling in each area is completed.

Further, the specific optimization objective of the particle swarm optimization model is shown in formula (6), including reducing power fluctuations and reducing new energy curtailment, and the constraints are shown in formula (8), including equipment output limitations and power fluctuation rates limit,

Among them, T _new represents the number of periods for calculating power fluctuation, λ _o,t is the purchase price of electric energy, λ _flu is the cost coefficient of power fluctuation loss, _PG,t is the actual output of new energy, and P _Go,t is the output capacity of new energy , _PGmax,t and _PGmin,t represent the upper and lower limits of the new energy equipment output during the t period, δ _t represents the power fluctuation of the new energy output during the t period, _δrise is the upper limit of the power increase rate, and _δdrop is the power reduction The lower limit of the rate, Δt is the time interval between two periods.

A new energy cluster consumption intelligent control system based on federated learning, including: a computer-readable storage medium and a processor;

The computer-readable storage medium is used to store executable instructions;

The processor is configured to read the executable instructions stored in the computer-readable storage medium, and execute the federated learning-based intelligent control method for new energy cluster consumption.

A non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the described intelligent control method for new energy cluster consumption based on federated learning is realized.

The present invention has following beneficial effects:

(1) The new energy cluster consumption intelligent control method based on federated learning in the present invention can establish multi-time-scale adjustable index sets, and consider the influence of source-load resource uncertainty factors to evaluate the control potential of cluster resources, so that energy companies According to the adjustable characteristics of different source-load resource clusters, orderly management strategies for different cluster source-load resources can be formulated to improve energy management efficiency;

(2) The new energy cluster consumption intelligent control method based on federated learning of the present invention can realize the localized utilization and protection of data without uploading to other control centers, and at the same time realize the update of model parameters. On the one hand, this method can effectively reduce The transmission process of the user's actual data is protected locally, which improves the privacy and security of the user data; on the other hand, this method only transmits parameters instead of direct data, and distributes the optimization calculation pressure of the control center to each sub-management center , which can reduce the communication cost and improve the update efficiency of optimized model parameters.

Description of drawings

Fig. 1 is the relationship diagram of adjustable resource response and incentive intensity under the influence of uncertainty factors in the present invention;

Fig. 2 is a parameter transfer process diagram of the federated learning method adopted by the present invention;

Fig. 3 is a flow chart of one embodiment of the federated learning-based new energy cluster consumption intelligent control method of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

As shown in Figure 3, the embodiment of the present invention provides an intelligent control method for new energy cluster consumption based on federated learning, including the establishment of multi-time scale index sets, the evaluation of scheduling potential considering the uncertainty of source-load resources, and the Intelligent scheduling, specifically including the following steps:

Step 1: Collect data of different new energy power generation equipment and power users, consider the uncertainty of source-load resource regulation, and establish a multi-time scale index set, which includes power generation and power consumption characteristic index sets of different time scales . The step 1 specifically includes the following steps:

Step 1.1: Determine the external factors affecting the dispatch potential uncertainty of new energy output resources, collect daily data, and establish a set of power generation characteristic indicators on different time scales.

The external influencing factors of intelligent regulation of new energy power generation resources mainly refer to weather factors, such as wind speed, solar radiation, temperature, humidity and other factors. Climatic factors such as rain, sunshine and snow will also have an impact on the power consumption of system equipment and power generation of power plants. The detailed indicators are shown in Table 1, and these indicators can be obtained through statistical calculation of daily power generation data.

Table 1. Power generation characteristic index set at different time scales

Step 1.2: Determine the internal influencing factors of the dispatch potential uncertainty of user demand response resources, collect data, and establish a set of electricity consumption characteristic indicators on different time scales.

Uncertainty in load model parameters is an important factor affecting response potential. For example, refined modeling for adjustable resources such as temperature-controlled loads and electric vehicles can more accurately obtain the response potential of user equipment, and then achieve accurate differential control. However, in many cases, the equivalent parameters of user equipment and electric vehicle charging Model parameters are difficult to measure directly, and as these parameters change with factors such as the environment, market, and user behavior, the model parameters will also change accordingly.

Therefore, the research on these load models should not be limited to the research on static load models, but should also consider the uncertainty of model parameters brought about by various influencing factors. For example: the aggregation model of air conditioner adjustable resources, the model parameters such as the equivalent heat capacity C of the aggregate, the equivalent thermal resistance R, and the energy efficiency ratio η are affected by various uncertain factors, and the adjustable resource clusters at each time node The equivalent model parameters of will also change. In addition, the factors influencing the adjustable potential of temperature control load include set temperature T _set , allowable temperature adjustment ΔT, regulation duration Δt, user water consumption m _n (water heater load), ambient temperature T _out , market electricity price TOU, etc.

The detailed indicators are shown in Table 2, and these indicators can be obtained through statistical calculation of daily power generation data.

Table 2. Electricity consumption characteristic index set at different time scales

Step 2: Based on the multi-time scale index set, considering the uncertainty of source and load resources, evaluate the potential of adjustable resources to participate in response scheduling, and obtain the potential evaluation results of adjustable resources. Described step 2 comprises the following steps:

Step 2.1: Consider the influence of uncertain factors mentioned in Step 1, and establish an assessment model for the response potential of adjustable resources.

As shown in Figure 1, the relationship between the response of adjustable resources and the intensity of incentives under the influence of uncertain factors, considering the influence of various uncertain factors in the evaluation of the response potential of adjustable resources, the linear region is generally considered to be represented by the black solid in Figure 1 The area surrounded by lines indicates that under a certain excitation strength in the linear area, the user response rate is not the only point, but changes within a possible area.

Neglecting the randomness of the response of adjustable resources in the dead zone and saturation zone, it mainly focuses on the modeling and solution of related parameters in the linear zone, including the inflection point r ₁ of the dead zone, the inflection point r ₂ of the saturation zone and its ordinate r ₃ . Due to the consideration of the influence of uncertain factors, it is impossible to use linear functions to model the linear region, but the response curve of the linear region is represented by a quadratic function:

η=r ₄ (δ-a)(δ-b) (1)

In the formula: r ₄ , a, b are related parameters of the quadratic model; η is the user response rate; δ is the incentive intensity.

Step 2.2: According to the known key parameters of the power generation equipment and user response model, the adjustable resource response potential evaluation model is converted into a solvable form.

If the key parameters of the model that a user participates in the response are known, that is, the inflection points (r ₁ , 0) and (r ₂ , r ₃ ), the user response load reduction rate can be obtained by the formula:

The r ₄ parameter in the formula is a random parameter considering the influence of uncertainty, and the random characteristics of the adjustable resource participating in the response process can be described through the random change of r ₄ . It is considered that the user's deterministic parameters r ₁ , r ₂ , and r ₃ have individual differences, while the impact of the random parameter r ₄ on the response potential of the same type of adjustable resources is similar. The random parameter r ₄ of a class of adjustable resource clusters with similar power consumption behavior can be analyzed, so as to obtain the response potential of cluster adjustable resources under the influence of uncertain factors.

Step 2.3: Through information mining of the daily data such as the user's power consumption capacity and production power consumption characteristics in the indicator set obtained in step 1, the definite parameters r1, r2, and r3 in the adjustable resource response potential evaluation model are obtained.

Considering that there are many indicators in the feature set, and some indicators have repeated feature information, the principal component analysis method is used to reduce the dimensionality of the indicator set obtained in step 1, and on this basis, the least squares fitting is used to mine the user-adjustable potential model Key parameters, and finally get the following relationship:

In the formula: b _1j , b _2j , and b _3j are the coefficients of the principal components of the key parameter influencing factors; U _j represents the principal component of the jth user’s electricity consumption characteristic index extracted by the principal component analysis method; a _ij is the jth The coefficient of the i-th index in the composition of principal components; x _i is the value of the i-th index. As the common characteristics of this type of users, the relationship in the formula can be directly used to obtain the determination parameters of general user response characteristics.

Step 2.4: According to the historical response data of user-adjustable resources, substituting into formula (2) can inversely calculate the value of the corresponding random parameter _r4 , thus forming a random parameter historical data set based on user historical response data.

Assuming that the random parameter r ₄ satisfies a normal distribution within a certain range, the box method can be used to determine the distribution robust fuzzy set of the random parameter:

标准差为

参数μ₀与σ₀为根据r₄的历史响应数据集进行点估计得到的估计值。Among them, r ₁ , r ₂ , r ₃ are known deterministic model parameters; r ₄ satisfies a normal distribution with certain rules, and the mean of the distribution is

standard deviation is

Parameters μ ₀ and σ ₀ are estimated values obtained by point estimation based on the historical response data set of r ₄ .

Step 2.5: Treat the obtained estimated value as a normal distribution parameter that r ₄ satisfies, so as to obtain the potential evaluation result of the adjustable resource participating in the response scheduling.

Among them, the mean reference value μ, the variance reference value σ ² , the mean fuzzy value μ _e and the variance fuzzy value σ _e are respectively model indicators describing the uncertainty parameters of user response potential. Uncertainty about the potential for business adjustment.

Step 3: According to the potential evaluation results of the adjustable resources, intelligently schedule the resources of the entire network based on federated learning. Described step 3 specifically comprises the following steps:

Step 3.1: Aiming at the optimal overall effect of new energy consumption, establish a particle swarm optimization model.

The specific optimization goal is shown in formula (6), including reducing power fluctuations and reducing power curtailment of new energy sources, and the constraints are shown in (8), including equipment output limitations and power fluctuation rate limitations;

Wherein, T _new represents the number of periods for calculating power fluctuations. In period t, λ _o,t is the purchase price of electric energy, and λ _flu is the cost coefficient of power fluctuation loss. P _G,t is the actual contribution of new energy. P _Go,t is the output capacity of new energy. _PGmax,t and _PGmin,t represent the upper and lower limits of the output of new energy equipment during the period t. Use _δt to represent the fluctuation of new energy output power in period t. δ _rise is the upper limit of the power increase rate, and δ _drop is the lower limit of the power decrease rate. Δt is the time interval between two periods.

Step 3.2: The general control center server uses the idea of federated learning to control the resources of the entire network. The specific implementation steps are as follows:

Step (a) The general control center server transmits the parameter w of the fixed particle swarm optimization algorithm model to each sub-regional center.

Step (b) Based on the potential evaluation probability results of adjustable resources participating in response scheduling obtained in step 2, workers in each sub-regional center obtain the power data of local adjustable resources participating in demand response. Using the source load power data information and the model parameter w from the general control center, the sub-regional center workers perform local conventional particle swarm optimization to calculate the new energy output or user load energy adjustment in each region.

Step (c): Each sub-regional center worker uses local data to perform multiple gradient descents locally, obtains an updated gradient g, and calculates a new parameter w and returns it to the general control center server.

w←w-α·g (8)

Among them, w is the particle swarm optimization model parameter, α is the specific weight coefficient update step size, and g is the gradient.

Step (d): The general control center server collects the new parameters w returned by the workers in each region (as shown in Figure 2), performs weighted average of the parameters, updates the parameters of the particle swarm model, and then sends them to the workers in each sub-region center.

Step (e): Repeat steps (b)-(d) until the controllable source-load resource scheduling in each area is completed.

The embodiment of the present invention also provides an intelligent control device for new energy cluster consumption based on federated learning, including:

The characteristic index set building module is used to collect data of different new energy power generation equipment and power users, and consider the uncertainty of source-load resource regulation to establish a multi-time scale index set. The multi-time scale index set includes different time scales of power generation and Electricity characteristic indicator set;

The potential evaluation module is configured to evaluate the potential of the adjustable resource to participate in the response scheduling based on the multi-time scale index set, considering the uncertainty of the source and load resources, and obtain the potential evaluation result of the adjustable resource;

The intelligent scheduling module is used to intelligently schedule the resources of the entire network based on federated learning based on the potential evaluation results of the adjustable resources.

Wherein, the characteristic index set establishment module includes a power generation characteristic index set establishment module and a power consumption characteristic index set establishment module;

The power generation characteristic index set building module is used to determine the external influence factors of the uncertainty of dispatch potential of new energy output resources, collect daily data, and establish power generation characteristic index sets of different time scales;

The external influencing factors of the dispatch potential uncertainty include weather factors and climate factors. The weather factors include wind speed, solar radiation, temperature, and humidity. The climate factors include rain, sunshine and snow. The characteristic index set is shown in Table 1, which is obtained through statistical calculation of daily power generation data:

Table 1. Power generation characteristic index set at different time scales

The establishment module of the electricity consumption characteristic index set is used to determine the internal influencing factors of the dispatch potential uncertainty of user demand response resources, collect data, and establish electricity consumption characteristic index sets of different time scales.

The internal influencing factors of the scheduling potential uncertainty include the equivalent heat capacity C of the polymer, the equivalent thermal resistance R, the energy efficiency ratio η, the set temperature T _set , the allowable temperature adjustment ΔT, the control duration Δt, and the user's water consumption m _n , ambient temperature T _out , market electricity price TOU, the power consumption characteristic index sets of different time scales are shown in Table 2, obtained through statistical calculation of daily power generation data:

Table 2. Electricity consumption characteristic index set at different time scales

The potential assessment module includes a potential assessment model building module and a potential assessment result acquisition module; wherein,

The potential evaluation model building module is used to establish an adjustable resource response potential evaluation model; considering the influence of various uncertain factors on the adjustable resource response potential evaluation, under a certain incentive strength in the linear region, the user response rate is not unique A point, but changes in a possible area;

Neglecting the randomness of the response of adjustable resources in the dead zone and saturation zone, model and solve the relevant parameters in the linear zone, specifically including the inflection point r ₁ of the dead zone, the inflection point r ₂ of the saturation zone and their ordinate r ₃ , and the response curve of the linear zone Represented by a quadratic function:

η=r ₄ (δ-a)(δ-b) (1)

In the formula: r ₄ , a, b are related parameters of the quadratic model; η is the user response rate; δ is the incentive intensity;

The potential evaluation model conversion module is used to convert the adjustable resource response potential evaluation model into a solvable form according to the known key parameters of the power generation equipment and user response model:

The r ₄ parameter in the formula is _a random parameter considering the influence of uncertainty. The random characteristics of adjustable resources participating in _the response process can be _described through the random change of r ₄ . individual differences, and the random parameter r ₄ has a similar impact on the response potential of the same type of adjustable resources, analyze the random parameter r ₄ of a class of adjustable resource clusters with similar electricity consumption behaviors, and thus obtain the uncertain Cluster adjustable resource response potential under the influence of factors;

Through information mining of the daily data in the multi-time-scale index set, the determined parameters r1, r2, and r3 in the adjustable resource response potential evaluation model are obtained:

The principal component analysis method is used to reduce the dimensionality of the multi-time scale index set, and on this basis, the least squares fitting is used to mine the key parameters of the user-adjustable potential model, and finally the following relationship is obtained:

In the formula: b _1j , b _2j , and b _3j are the coefficients of the principal components of the key parameter influencing factors; U _j represents the principal component of the jth user’s electricity consumption characteristic index extracted by the principal component analysis method; a _ij is the jth The coefficient of the i-th index in the composition of principal components; x _i is the value of the i-th index;

According to the historical response data of user-adjustable resources, substitute into formula (2) to find the corresponding r ₄ value, thus forming a random parameter historical data set based on user historical response data; assuming that the random parameter satisfies a normal distribution within a certain range, The distribution robust fuzzy set of this random parameter is thus determined by the box method:

标准差为

参数μ₀与σ₀为根据r₄的历史响应数据集进行点估计得到的概率估计值；Among them, r ₁ , r ₂ , r ₃ are known deterministic model parameters; r ₄ is a normal distribution that satisfies certain rules, and the mean of the distribution is

standard deviation is

Parameters μ ₀ and σ ₀ are probability estimates obtained by point estimation based on the historical response data set of r ₄ ;

The potential evaluation result acquisition module is used to treat the obtained probability estimate as a normal distribution parameter satisfied by r ₄ , so as to obtain the potential evaluation result of the adjustable resource participating in the response scheduling:

Among them, the mean reference value μ, the variance reference value σ ² , the mean fuzzy value μ _e and the variance fuzzy value σ _e are respectively model indicators describing the uncertainty parameters of user response potential. Uncertainty about the potential for business adjustment.

The intelligent scheduling module includes a particle swarm optimization model establishment module and a resource regulation module of the whole network;

The particle swarm optimization model building module is used to establish the particle swarm optimization model with the goal of optimizing the overall effect of new energy consumption:

The specific optimization objective is shown in formula (6), including reducing power fluctuations and reducing new energy curtailment. The constraints are shown in formula (8), including equipment output limitations and power fluctuation rate limitations.

Among them, T _new represents the number of periods for calculating power fluctuation, λ _o,t is the purchase price of electric energy, λ _flu is the cost coefficient of power fluctuation loss, _PG,t is the actual output of new energy, and P _Go,t is the output capacity of new energy , _PGmax,t and _PGmin,t represent the upper and lower limits of the new energy equipment output during the t period, δ _t represents the power fluctuation of the new energy output during the t period, _δrise is the upper limit of the power increase rate, and _δdrop is the power reduction The lower limit of the rate, Δt is the time interval between two periods;

The resource control module of the whole network includes the general control center server and the regional center worker;

The general control center server is used to use the idea of federated learning to control the resources of the whole network, including:

Step (a) the general control center server transmits the parameter w of the fixed particle swarm optimization algorithm model to each sub-regional center;

Step (b) Based on the potential evaluation results of the adjustable resources, the workers in each sub-regional center obtain the power data of the local adjustable resources participating in demand response, and use the source-load power data information and the model parameters transmitted from the general control center w, sub-regional center workers perform local conventional particle swarm optimization respectively, and calculate the new energy output of each region or the adjustment amount of user load energy consumption;

Step (c) Each sub-regional center worker uses local data to perform multiple gradient descents locally, obtains the updated gradient g, and calculates the new parameter w and returns it to the general control center server.

w←w-α·g (9)

Among them, w is the particle swarm optimization model parameter, α is the specific weight coefficient update step size, and g is the gradient;

Step (d) The general control center server collects the new parameters w returned by the workers in each area, performs weighted average on the parameters, updates the parameters of the particle swarm model, and then sends them to the workers in each sub-area center;

Step (e) Repeat steps (b)-(d) until the controllable source-load resource scheduling in each area is completed.

Considering the influence of uncertain factors, the present invention realizes the potential evaluation of cluster control of new energy equipment and users, and based on the idea of federated learning, protects the data privacy of each region in the subsequent resource control process and improves the overall intelligent control. safety.

Another aspect of the present invention provides a new energy cluster consumption intelligent control system based on federated learning, including: a computer-readable storage medium and a processor;

The computer-readable storage medium is used to store executable instructions;

The processor is configured to read the executable instructions stored in the computer-readable storage medium, and execute the federated learning-based intelligent control method for new energy cluster consumption described in the first aspect.

Another aspect of the present invention provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the new energy cluster consumption intelligence based on federated learning described in the first aspect is realized. control method.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be Any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention shall fall within the protection scope of the claims of the present invention.

Claims (11)

1. A new energy cluster consumption intelligent control method based on federal learning is characterized in that: the method comprises the following steps:

collecting data of different new energy power generation equipment and power users, and establishing a multi-time scale index set, wherein the multi-time scale index set comprises power generation and power utilization characteristic index sets with different time scales;

based on the multi-time scale index set, evaluating the potential of the adjustable resource participating in response scheduling to obtain a potential evaluation result of the adjustable resource;

and aiming at the potential evaluation result of the adjustable resources, intelligently scheduling the resources of the whole network based on federal learning.

2. The new energy cluster digestion intelligent control method based on federal learning according to claim 1, characterized in that: the method comprises the following steps of collecting data of different new energy power generation equipment and power users, establishing a multi-time scale index set, wherein the multi-time scale index set comprises power generation and power utilization characteristic index sets with different time scales, and specifically comprises the following steps:

determining external influence factors of scheduling potential uncertainty of new energy output resources, collecting daily data, and establishing power generation characteristic index sets of different time scales;

the external influence factors of the scheduling potential uncertainty comprise weather factors and climate factors, the weather factors comprise wind speed, solar radiation degree, temperature and humidity, the climate factors comprise rain, sunny days and snow, and the power generation characteristic index sets of different time scales are obtained through statistical calculation of daily power generation data;

determining internal influence factors of scheduling potential uncertainty of user demand response resources, collecting data, and establishing power utilization characteristic index sets with different time scales;

the internal influence factors of the scheduling potential uncertainty comprise the equivalent heat capacity C, the equivalent heat resistance R, the energy efficiency ratio eta and the set temperature T of the polymer _set Allowable temperature adjustment quantity delta T, regulation duration time delta T and user water consumption m _n Ambient temperature T _out And the electricity utilization characteristic index sets of different time scales are obtained by daily electricity generation data statistics and calculation.

3. The new energy cluster consumption intelligent control method based on federal learning according to claim 1, characterized in that: based on the multi-time scale index set, the potential of the adjustable resource participating in response scheduling is evaluated to obtain a potential evaluation result of the adjustable resource, and the method specifically comprises the following steps:

establishing an adjustable resource response potential evaluation model:

wherein r is ₁ 、r ₂ 、r ₃ Is a known deterministic model parameter; r is ₄ To satisfy a regular normal distribution, the mean value of the distribution is

Standard deviation of

Parameter mu ₀ And σ ₀ Is according to r ₄ Carrying out point estimation on the historical response data set to obtain a probability estimation value;

the resulting probability estimate is taken as r ₄ And (3) meeting the normal distribution parameters, thereby obtaining a potential evaluation result of the adjustable resource participating in response scheduling:

wherein, the mean value reference value mu and the variance reference value sigma ² Mean fuzzy value μ _e And variance ambiguity value σ _e The model indexes are model indexes for describing uncertainty parameters of user response potential, and the 4 indexes comprehensively reflect the uncertainty of the adjustable potential of the adjustable resource participating response service.

4. The intelligent new energy cluster absorption regulation and control method based on federal learning according to claim 1, which is characterized in that: aiming at the potential evaluation result of the adjustable resources, intelligently scheduling the resources of the whole network based on federal learning, specifically comprising the following steps:

establishing a particle swarm optimization model by taking the optimal total effect of new energy consumption as a target:

the specific optimization target is shown in formula (6) and comprises reducing power fluctuation and reducing new energy power abandon, the limiting conditions are shown in formula (8) and comprise equipment output limit and power fluctuation rate limit,

wherein T is _new Indicating the number of periods during which the power fluctuation is calculated, λ _o,t For purchase price of electric energy, λ _flu For power fluctuation loss cost factor, P _G,t For actual output of new energy, P _Go,t Capability of new energy to output, P _Gmax,t 、P _Gmin,t Represents the upper and lower limit of the new energy equipment output in the t period, delta _t Representing the new energy output power fluctuation, delta, over a period of t _rise As an upper limit of the power increase rate, δ _drop To lower the power reduction rate, Δ t is the time interval between two periods.

5. The new energy cluster consumption intelligent control method based on federal learning according to claim 4, characterized in that: the method comprises the following steps that a general control center server uses a federal learning idea to regulate and control resources of the whole network, and specifically comprises the following steps:

the method comprises the following steps that (a) a master control center server transmits a parameter w of a fixed particle swarm optimization algorithm model to each regional center;

step (b) based on the potential evaluation result of the adjustable resources, the worker in each regional center obtains power data of local adjustable resources participating in demand response, local conventional particle swarm optimization is respectively carried out on the worker in each regional center by utilizing the source charge power data information and the model parameter w transmitted from the total regulation center, and the adjustment quantity of new energy output or user load energy consumption in each region is calculated;

step (c), each regional center worker locally performs multiple gradient descent by using local data to obtain an updated gradient g, calculates a new parameter w and returns the new parameter w to the general control center server,

w←w-α·g (8)

wherein w is a particle swarm optimization model parameter, alpha is a specific weight coefficient updating step length, and g is a gradient;

the master control center server collects new parameters w returned by each regional worker, carries out weighted average on the parameters, updates particle swarm model parameters, and then sends the parameters to each regional center worker;

and (e) repeating the steps (b) to (d) until the controllable source load resource scheduling in each area is completed.

6. The utility model provides a new forms of energy cluster consumption intelligent control device based on federal study which characterized in that includes:

the characteristic index set establishing module is used for collecting data of different new energy power generation equipment and power users and establishing a multi-time scale index set, wherein the multi-time scale index set comprises power generation and power utilization characteristic index sets with different time scales;

the potential evaluation module is used for evaluating the potential of the adjustable resource participating in response scheduling based on the multi-time scale index set to obtain a potential evaluation result of the adjustable resource;

and the intelligent scheduling module is used for intelligently scheduling the resources of the whole network based on federal learning aiming at the potential evaluation result of the adjustable resources.

7. The intelligent new energy cluster digestion regulation and control device based on federal learning of claim 6, wherein the characteristic index set establishment module comprises a power generation characteristic index set establishment module and a power utilization characteristic index set establishment module;

the power generation characteristic index set establishing module is used for determining external influence factors of scheduling potential uncertainty of the new energy output resources, collecting daily data and establishing power generation characteristic index sets with different time scales;

the external influence factors of the scheduling potential uncertainty comprise weather factors and climate factors, the weather factors comprise wind speed, solar radiation degree, temperature and humidity, the climate factors comprise rain, sunny days and snow, and the power generation characteristic index sets of different time scales are obtained through statistical calculation of daily power generation data;

the power utilization characteristic index set establishing module is used for determining internal influence factors of scheduling potential uncertainty of user demand response resources, collecting data and establishing power utilization characteristic index sets with different time scales.

The internal influence factors of the scheduling potential uncertainty comprise the equivalent heat capacity C, the equivalent heat resistance R, the energy efficiency ratio eta and the set temperature T of the polymer _set Allowable temperature adjustment quantity delta T, regulation duration time delta T and user water consumption m _n Ambient temperature T _out And the market electricity price TOU, wherein the electricity utilization characteristic index sets of different time scales are obtained by statistical calculation of daily electricity generation data.

8. The intelligent regulation and control device of new energy cluster consumption based on federal learning of claim 7, comprising: the potential evaluation module comprises a potential evaluation model establishing module and a potential evaluation result obtaining module; wherein,

the potential evaluation model establishing module is used for establishing an adjustable resource response potential evaluation model:

wherein r is ₁ 、r ₂ 、r ₃ Is a known deterministic model parameter; r is ₄ To satisfy a regular normal distribution, wherein the mean of the distribution is

Standard deviation of

Parameter mu ₀ And σ ₀ Is according to r ₄ Carrying out point estimation on the historical response data set to obtain a probability estimation value;

a potential evaluation result acquisition module for considering the obtained probability estimation value as r ₄ And (3) satisfying the normal distribution parameters, thereby obtaining a potential evaluation result of the adjustable resource participating in response scheduling:

wherein, the mean value reference value mu and the variance reference value sigma ² Mean fuzzy value μ _e And variance ambiguity value σ _e The model indexes are model indexes for describing uncertainty parameters of user response potential, and the 4 indexes comprehensively reflect the uncertainty of the adjustable potential of the adjustable resource participating response service.

9. The intelligent new energy cluster digestion regulation and control device based on federal learning of claim 7, wherein the intelligent scheduling module comprises a particle swarm optimization model building module and a whole network resource regulation and control module;

and the particle swarm optimization model establishing module is used for establishing a particle swarm optimization model by taking the optimal total effect of new energy consumption as a target.

10. The intelligent new energy cluster digestion regulation and control device based on federal learning of claim 9,

the whole network resource regulation and control module comprises a main regulation and control center server and a regional center worker;

the general regulation and control center server is used for using the federal learning idea to regulate and control resources of the whole network, and specifically comprises the following steps:

the method comprises the following steps that (a) a master control center server transmits a parameter w of a fixed particle swarm optimization algorithm model to each regional center;

step (b) based on the potential evaluation result of the adjustable resources, the worker in each regional center obtains power data of local adjustable resources participating in demand response, local conventional particle swarm optimization is respectively carried out on the worker in each regional center by utilizing the source charge power data information and the model parameter w transmitted from the total regulation center, and the adjustment quantity of new energy output or user load energy consumption in each region is calculated;

step (c), each regional center worker performs gradient descent for multiple times locally by using local data to obtain an updated gradient g, calculates a new parameter w and returns the new parameter w to the general control center server,

w←w-α·g (9)

w is a particle swarm optimization model parameter, alpha is a specific weight coefficient updating step length, and g is a gradient;

and (d) the master control center server collects new parameters w returned by each regional worker, performs weighted average on the parameters, updates the particle swarm model parameters, and then sends the parameters to each regional center worker.

And (e) repeating the steps (b) to (d) until the controllable source load resource scheduling in each area is completed.

11. The Federal learning-based intelligent new energy cluster absorption regulation and control device as claimed in claim 9, wherein the specific optimization objective of the particle swarm optimization model is shown in formula (6) and includes reducing power fluctuation and reducing new energy electricity abandonment, and the limiting conditions are shown in formula (8) and include equipment output limit and power fluctuation rate limit,

wherein T is _new Indicating the number of periods during which the power fluctuation is calculated, lambda _o,t For purchase price of electric energy, λ _flu For power fluctuation loss cost factor, P _G,t For actual contribution of new energy, P _Go,t For new energy possible capacity, P _Gmax,t 、P _Gmin,t Represents the upper and lower limit of the new energy equipment output in the t period, delta _t Represents t timeNew energy output power fluctuation of the segment, delta _rise As an upper limit of the power increase rate, δ _drop To lower the power reduction rate, Δ t is the time interval between two periods.

Images