CN115935285A - Multi-element time series anomaly detection method and system based on mask map neural network model - Google Patents
Multi-element time series anomaly detection method and system based on mask map neural network model Download PDFInfo
- Publication number
- CN115935285A CN115935285A CN202211405450.XA CN202211405450A CN115935285A CN 115935285 A CN115935285 A CN 115935285A CN 202211405450 A CN202211405450 A CN 202211405450A CN 115935285 A CN115935285 A CN 115935285A
- Authority
- CN
- China
- Prior art keywords
- time sequence
- node
- sliding window
- value
- time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a multivariate time series anomaly detection method and a multivariate time series anomaly detection system based on a mask map neural network model, wherein the method comprises the following steps: carrying out data preprocessing operation on the multivariate time sequence; inputting the preprocessed multivariate time sequence; carrying out proportional mask operation on time sequence data in the sliding window; establishing a graph structure by utilizing the similarity of each node embedded expression vector; aggregating neighbor node information for each graph node according to the time sequence in the sliding window and by using the current graph structure, learning the characteristic representation of the sliding window, and predicting the value of the next timestamp by using the characteristic; carrying out generative confrontation training on the predicted value and the true value, calculating the loss of a generator and the loss of a discriminator, and updating the model; and calculating the abnormal score of each node according to the difference between the predicted value and the true value, and judging the time point with the abnormal score higher than the threshold value as an abnormal time point to obtain an abnormal analysis result. The invention is helpful for accurately detecting the abnormality of the multivariate time variable data.
Description
Technical Field
The invention belongs to the technical field of data mining, and particularly relates to a multivariate time series anomaly detection method and system based on a mask map neural network model.
Background
In order to ensure the normal operation of the network system, operation and maintenance personnel need to monitor a large amount of data at any time. The data is from a plurality of interrelated monitoring devices that are generated as the system operates, forming a multivariate time series. The time series abnormity detection is a bottom core work of intelligent operation and maintenance, and the aim of the time series abnormity detection is to analyze the change of a time series and find out abnormal outliers or abnormal columns from a large amount of data, namely, the abnormal conditions of software and hardware services of a system. With the large-scale and complicated service system, the application of the internet of things and sensors is gradually expanded, and the detection of faults and the guarantee of safety through monitoring are very important.
The unsupervised anomaly detection has wider application range because the application environment often lacks enough anomaly markers under common conditions and the variation of the anomaly is irregular. Intuitively, the task can be accomplished without the help of annotation data by setting a threshold and identifying data outside of this range as anomalous data. The following common unsupervised abnormality detection methods exist at present: statistical learning-based methods such as principal component analysis, distance-based clustering, and density-based clustering, but these methods require a priori knowledge about anomalies; machine learning-based methods such as random forests, isolated forests, and single classification support vector machines have also been applied to anomaly detection, but such methods are relatively simple in fitting the distribution of anomalous data and are not sufficient to accurately detect anomalies in multivariate time data; in addition, the method based on deep learning has the defects that the network structure is complex, the calculation amount is large, the real-time performance is poor, and the like.
In view of the above, it is necessary to design a new multivariate time series anomaly detection method to solve the above problems.
Disclosure of Invention
The invention mainly aims to provide a multivariate time series abnormality detection method and a multivariate time series abnormality detection system based on a mask map neural network model, which are beneficial to accurately detecting the abnormality of multivariate time variable data.
In order to achieve the above object, the present invention provides a multivariate time series anomaly detection method based on a mask map neural network model, which comprises the following steps:
s1, performing data preprocessing operation on the multi-element time sequence;
s2, inputting the preprocessed multivariate time sequence, intercepting a fixed length, namely a sliding window, of each sensor according to a time sequence, and sliding with a fixed step length to obtain a final data set, wherein each sensor is a graph node, and the corresponding time sequence is a node characteristic; carrying out proportional mask operation on time sequence data in the sliding window;
s3, initializing a node embedded expression vector for each sensor, updating, and establishing a graph structure by using the similarity of each node embedded expression vector;
s4, aggregating neighbor node information for each graph node according to the time sequence in the sliding window and by using the current graph structure, learning the characteristic representation of the sliding window, and predicting the value of the next time stamp by using the characteristic;
s5, performing generative confrontation training on the predicted value and the true value, calculating the loss of a generator and the loss of a discriminator, and updating the model;
and S6, calculating the abnormal score of each node through the difference between the predicted value and the true value in the testing stage, selecting the point with the highest score as a threshold, and judging the time point with the abnormal score higher than the threshold as an abnormal time point to obtain the final abnormal analysis result.
A further development of the invention is that step S1 comprises the following steps:
processing the missing value, and filling 0 in the missing part;
sampling the time sequence, and selecting a group of data at intervals of a fixed time period;
and (4) normalizing the numerical values of the whole time sequence of each sensor, and compressing the data to be between 0 and 1.
A further development of the invention is that step S2 comprises the following steps:
intercepting a fixed length, namely a sliding window, of each sensor according to a time sequence, and sliding in a fixed step length until the end of time to obtain a processed data set;
specifying a maximum sequence length of the continuous mask to ensure continuity of the time sequence; the maximum continuous mask sequence length is half of the total number of the shielding time points;
carrying out proportional random mask on the time sequence in the sliding window, and setting the value of the sensor at the time point of the mask to be 0; and if the length of the continuous mask sequence exceeds the length of the maximum mask sequence, performing mask operation according to the length of the maximum mask sequence.
A further development of the invention is that step S3 comprises the following steps: initializing a node embedded expression vector for each sensor;
updating the node embedded expression vector by using an encoder, converting the randomly initialized node embedded expression vector into a dense vector with the same dimension, and keeping the encoder parameter updated;
in order to control the number of the neighbor nodes to improve the operation efficiency and avoid overfitting, the number of the maximum neighbor nodes is set to control the number of the edges;
and constructing a graph structure according to the cosine similarity of the node embedded expression vector.
A further development of the invention is that step S4 comprises the following steps:
aggregating neighbor node information for each graph node by using an attention mechanism according to the time sequence in the sliding window and the current graph structure, and learning out the characteristic representation of the sliding window;
inputting the characteristic representation of the sliding window into the multilayer fully-connected network, and predicting the value of the next time stamp of each graph node.
A further development of the invention is that step S5 comprises the following steps:
the generator is a graph model based on an attention mechanism, the loss of the generator is calculated, and the generator is updated;
and the GAN model based on the discriminator is used for splicing the real value of the next time stamp after the time sequence in the sliding window to form a real time sequence, splicing the predicted value of the next time stamp after the time sequence in the sliding window to form a predicted time sequence, inputting the real time sequence and the predicted time sequence into the discriminator, calculating the loss of the discriminator and updating the discriminator.
A further development of the invention is that step S6 comprises the following steps:
performing no masking operation on the time sequence in the sliding window, and calculating the predicted value of each node;
calculating an abnormal score of each node through the difference between the predicted value and the real value;
to reduce the effect of the relative sizes of the different sequences, the present invention uses a smoothing function when computing the anomaly score: subtracting the average value of each sequence and dividing by the standard deviation, and selecting the highest score as a threshold value;
and judging the time point with the abnormal score higher than the threshold value as an abnormal time point to obtain a final abnormal analysis result.
In order to achieve the above object, the present invention further provides a multivariate time series anomaly detection system based on a mask map neural network model, which performs any one of the methods described above.
The invention has the following beneficial effects: the invention adopts an unsupervised method to solve the problems of difficult data annotation and unbalance; by combining time sequence mask operation and a graph structure, the learning capacity of a graph model is enhanced, and abnormal nodes and nodes influenced by the abnormal nodes are favorably positioned; the method has good flexibility and expansibility, and further improves the prediction precision or the operation efficiency by adjusting the network parameters more suitable for a certain working environment; experiments prove that the model has better performance in the aspect of anomaly detection than the most advanced anomaly detection technology. The results on the rich data set also prove that the anomaly detection method provided by the invention has universality.
Drawings
FIG. 1 is a schematic overall flow chart provided according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a masking operation according to an embodiment of the present invention;
FIG. 3 is a flow chart of a structure of a build graph provided according to an embodiment of the present invention;
fig. 4 is a schematic diagram of an anomaly detection process according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
It should be emphasized that in describing the present invention, various formulas and constraints are identified with consistent labels, but the use of different labels to identify the same formula and/or constraint is not precluded and is provided for the purpose of more clearly illustrating the features of the present invention.
The embodiment of the invention discloses a multivariate time series anomaly detection method based on a mask map neural network model, which mainly comprises the following steps:
s1, performing data preprocessing operation on the multivariate time sequence;
s2, inputting the preprocessed multi-element time sequence, intercepting a fixed length, namely a sliding window, for each sensor according to the time sequence, and sliding with a fixed step length to obtain a final data set, wherein each sensor is a graph node, and the corresponding time sequence is a node characteristic; carrying out proportional mask operation on time sequence data in the sliding window;
s3, initializing a node embedded expression vector for each sensor, updating, and establishing a graph structure by using the similarity of each node embedded expression vector;
s4, aggregating neighbor node information for each graph node according to the time sequence in the sliding window and by using the current graph structure, learning the characteristic representation of the sliding window, and predicting the value of the next time stamp by using the characteristic;
s5, performing generative confrontation training on the predicted value and the true value, calculating the loss of a generator and the loss of a discriminator, and updating a model;
and S6, calculating the abnormal score of each node through the difference between the predicted value and the true value in the testing stage, selecting the point with the highest score as a threshold, and judging the time point with the abnormal score higher than the threshold as an abnormal time point to obtain the final abnormal analysis result.
The method of the present invention is described in detail below with reference to fig. 1 to 4.
In step S1, further comprising:
first, missing value processing is performed on the data, and the missing portion is filled with 0.
Secondly, on the premise of not influencing the abnormal detection effect, in order to save training resources, the invention samples the time sequence, selects a group of data every other fixed time period, and represents the characteristics of the time period by the median of each sensor value.
And finally, performing normalization processing on the numerical value of the whole time sequence of each sensor, and compressing the data to 0-1 so as to accelerate the convergence rate of model training:
s2, inputting the preprocessed multi-element time sequence, intercepting a fixed length, namely a sliding window, for each sensor according to the time sequence, and sliding with a fixed step length to obtain a final data set, wherein each sensor is a graph node, and the corresponding time sequence is a node characteristic; and performing a scale mask operation on the time sequence data in the sliding window.
The specific method for carrying out the proportional mask operation on the time sequence data in the sliding window comprises the following steps:
firstly, a fixed length, namely a sliding window, is intercepted from each sensor according to a time sequence, and the sliding window is slid by a fixed step length until the end of time, so that a processed data set is obtained. In this embodiment, the multivariate time variable can be expressed as X = { xt } T e,
and n is the number of sensors. Setting the size of the sliding window to w, and selecting data in w consecutive time stamps of each sensor, such as ^ H>
And sliding in a fixed step length until the end of time, and recording data in the sliding window and data corresponding to the next timestamp to obtain a processed data set.
Second, a maximum sequence length of the continuous mask is specified to ensure continuity of the time sequence. The maximum continuous mask sequence length is half of the number of total masking time points.
Finally, the time series within the sliding window is randomly masked in proportion, and the value of the sensor at the time point of the masking is set to 0. And if the length of the continuous mask sequence exceeds the length of the maximum mask sequence, performing mask operation according to the length of the maximum mask sequence. In the embodiment of the invention, the time sequence in the sliding window is randomly removed with some parts according to a certain proportion. For example, when a 20% mask is taken for a sliding window input of length 20, the present invention randomly sets four time nodes to 0. The aim of the invention is to be able to generate a prediction value in the case of masking, and therefore, in order not to disrupt the continuity of the time series in a large range, the invention will specify the maximum sequence length of the masking operation. In general, the maximum contiguous mask length will not exceed half the masking rate. That is, the longest continuous mask length in the above example is 2.
And S3, initializing a node embedded expression vector for each sensor, updating, and establishing a graph structure by using the similarity of each node embedded expression vector. The step of establishing the graph structure comprises the following steps:
initializing a node embedded expression vector for each sensor;
updating the node embedded expression vector by using an encoder, converting the randomly initialized node embedded expression vector into a dense vector with the same dimension, and keeping the encoder parameter updated;
in order to control the number of neighbor nodes to improve the operation efficiency and avoid overfitting, the number of the maximum neighbor nodes is set to control the number of edges;
and constructing a graph structure according to the cosine similarity of the node embedded expression vector.
In this embodiment, as the training proceeds, the node-embedded representation vector is continuously updated, and a graph model most suitable for prediction is trained by using the node-embedded representation vector. The present invention uses node-embedded representation vectors as a means to represent each time series feature, mapping it to a high-dimensional vector. The vectors are initialized randomly, the invention hopes that the embedding can represent the characteristics between the nodes as accurately as possible, therefore, the embedded expression vectors of the nodes initialized randomly are updated by using an encoder, the embedded expression vectors of the nodes initialized randomly are converted into dense vectors with the same dimension, the encoder keeps the updating of the parameters of the embedded expression vectors during model training, and the vectors coded by the encoder are new embedded expression vectors of the nodes. Representing node embedded representation vectors as V i ∈R d And updating the graph structure by utilizing cosine similarity between the graph structure and the graph structure. V 1 ，V 2 Is a node v 1 ，v 2 The embedded expression vector of (2), then the node similarity R n (v 1 ，v 2 ) Comprises the following steps:
in order to control the number of the neighbor nodes to improve the operation efficiency and avoid overfitting, the invention adopts a strategy of setting the maximum number of the neighbor nodes in advance to control the number of the edges. For example, when the number of the maximum neighbor nodes is 20, the similarity between other nodes and the node a is sorted from high to low, and then the first 20 nodes are selected to establish a relationship with the node a.
And S4, aggregating neighbor node information for each graph node by using the current graph structure according to the time sequence in the sliding window, and predicting the value of the next timestamp. The step of predicting the value of the next timestamp comprises:
aggregating neighbor node information for each graph node by using an attention mechanism according to the time sequence in the sliding window and the current graph structure, and learning out the characteristic representation of the sliding window;
inputting the characteristic representation of the sliding window into the multilayer fully-connected network, and predicting the value of the next time stamp of each graph node.
In this embodiment, aggregating information about nodes and their correlation with their neighbors requires calculating attention coefficients α between graph nodes i，j . The attention coefficient represents the importance of the feature of node j to node i. To calculate alpha i，j Firstly, the influence of two nodes is considered simultaneously, and the attention value epsilon between the two nodes is calculated i，j ：
Where LeakyReLU is a non-linear activation function, a is a learning coefficient vector of the attention mechanism,
representing connections, W is a trainable linear transformation matrix with each node sharing weights. By embedding nodes to represent vector V i Characteristic of the current time>
In combination, a more comprehensive calculation of the attention factor can be performed. In aggregating neighbor information, the attention of all neighbors of each node needs to be normalized. The normalized attention weight is the aggregation coefficient α i，j ：
Obtaining an aggregate representation of each node by an aggregation coefficient
The aggregated representation of all nodes is then input into the multi-layer fully-connected network to obtain a predicted value for each node. The whole prediction process is denoted as G ():
and S5, carrying out generative confrontation training on the predicted value and the true value, calculating the loss of the generator and the loss of the discriminator, and updating the model. The method for generating the antagonistic training comprises the following steps:
the generator is a graph model based on an attention mechanism, the loss of the generator is calculated, and the generator is updated;
and the GAN model based on the discriminator is used for splicing the real value of the next time stamp after the time sequence in the sliding window to form a real time sequence, splicing the predicted value of the next time stamp after the time sequence in the sliding window to form a predicted time sequence, inputting the real time sequence and the predicted time sequence into the discriminator, calculating the loss of the discriminator and updating the discriminator.
For the specific embodiment, the key point of the graph structure model of the invention is to learn the data distribution of a normal time series, so that the data distribution needs to be trained, and the predicted value of the next timestamp is as close to the true value as possible. That is, when training in a countervailing manner, it is desirable to have the model generate as much spurious data as possible, thereby confusing the discriminators. The input through the time window is represented as: { x) 1 ，...，x t-1 Then x should be derived by graph-based model G () t ：
The generator is updated according to the generator losses.
The present invention uses the underlying GAN model as a discriminator. The discriminator needs to discriminate the data generated by the model from the real data. To allow for a deeper level of correlation, the present invention inputs the time series into the discriminator. And splicing the real value of the next time stamp after the time sequence in the sliding window to form a real time sequence, splicing the predicted value of the next time stamp after the time sequence in the sliding window to form a predicted time sequence, and inputting the real time sequence and the predicted time sequence into the discriminator. Let the discriminator be denoted D (), let x t In order to be the true data,
to predict data, the input can then be expressed as:
the discriminator loss is as follows:
and updating the discriminator according to the discriminator loss.
And S6, calculating the abnormal score of each node through the difference between the predicted value and the true value in the testing stage, selecting the point with the highest score as a threshold, and judging the time point with the abnormal score higher than the threshold as an abnormal time point to obtain the final abnormal analysis result. The method for detecting the abnormity of the test stage comprises the following steps:
performing no masking operation on the time sequence in the sliding window, and calculating the predicted value of each node;
calculating an abnormal score of each node through the difference between the predicted value and the real value;
to reduce the effect of the relative sizes of the different sequences, the present invention uses a smoothing function when computing the anomaly score: subtracting the average value of each sequence and dividing by the standard deviation, and selecting the highest score as a threshold value;
and judging the time point with the abnormal score higher than the threshold value as an abnormal time point to obtain a final abnormal analysis result.
For the specific embodiment, in the testing stage, the masking operation is not carried out to calculate the predicted value, and the abnormal score of each node is calculated through the difference between the predicted value and the real value. And selecting the index with the highest abnormal score as the basis for judging the abnormality at the time point. To reduce the effect of the relative sizes of the different sequences, the present invention uses a smoothing function: the mean of each sequence was subtracted and divided by the standard deviation:
ADS t ＝max{loss n } n∈N
and taking the moment when the abnormal score exceeds the threshold value as an abnormal time point to obtain a final abnormal analysis result.
Based on the above inventive concept, the present invention further provides a multivariate time series anomaly detection system based on a mask map neural network model, which includes at least one computing device, where the computing device includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and when the computer program is loaded into the processor, the multivariate time series anomaly detection method based on the mask map neural network model can be implemented.
The invention adopts an unsupervised method to solve the problems of difficult data annotation and unbalance; by combining time sequence mask operation and a graph structure, the learning capacity of a graph model is enhanced, and abnormal nodes and nodes influenced by the abnormal nodes are favorably positioned; the method has good flexibility and expansibility, and further improves the prediction precision or the operation efficiency by adjusting the network parameters more suitable for a certain working environment; experiments prove that the model has better performance in the aspect of anomaly detection than the most advanced anomaly detection technology. The results on the rich data set also prove that the anomaly detection method provided by the invention has universality.
Although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for elements thereof without departing from the spirit and scope of the present invention.
Claims (8)
1. A multivariate time series anomaly detection method based on a mask map neural network model is characterized by comprising the following steps:
s1, performing data preprocessing operation on the multivariate time sequence;
s2, inputting the preprocessed multi-element time sequence, intercepting a fixed length, namely a sliding window, for each sensor according to the time sequence, and sliding with a fixed step length to obtain a final data set, wherein each sensor is a graph node, and the corresponding time sequence is a node characteristic; carrying out proportional mask operation on time sequence data in the sliding window;
s3, initializing a node embedded expression vector for each sensor, updating, and establishing a graph structure by using the similarity of each node embedded expression vector;
s4, aggregating neighbor node information for each graph node according to the time sequence in the sliding window and by using the current graph structure, learning the characteristic representation of the sliding window, and predicting the value of the next time stamp by using the characteristic;
s5, performing generative confrontation training on the predicted value and the true value, calculating the loss of a generator and the loss of a discriminator, and updating the model;
and S6, calculating the abnormal score of each node through the difference between the predicted value and the true value in the testing stage, selecting the point with the highest score as a threshold, and judging the time point with the abnormal score higher than the threshold as an abnormal time point to obtain the final abnormal analysis result.
2. The method of claim 1, wherein: the step S1 includes the steps of:
processing the missing value, and filling 0 in the missing part;
sampling the time sequence, and selecting a group of data every other fixed time period;
and (4) normalizing the numerical values of the whole time sequence of each sensor, and compressing the data to be between 0 and 1.
3. The method of claim 2, wherein: the step S2 includes the steps of:
intercepting a fixed length, namely a sliding window, of each sensor according to a time sequence, and sliding in a fixed step length until the end of time to obtain a processed data set;
specifying a maximum sequence length of the continuous mask to ensure continuity of the time sequence; the maximum continuous mask sequence length is half of the total number of the shielding time points;
carrying out proportional random mask on the time sequence in the sliding window, and setting the value of the sensor at the time point of the mask to be 0; and if the length of the continuous mask sequence exceeds the length of the maximum mask sequence, performing mask operation according to the length of the maximum mask sequence.
4. The method of claim 3, wherein: step S3 includes the following steps: initializing a node embedded expression vector for each sensor;
updating the node embedded expression vector by using an encoder, converting the randomly initialized node embedded expression vector into a dense vector with the same dimension, and keeping the encoder parameter updated;
in order to control the number of neighbor nodes to improve the operation efficiency and avoid overfitting, the number of the maximum neighbor nodes is set to control the number of edges;
and constructing a graph structure according to the cosine similarity of the node embedded expression vector.
5. The method of claim 4, wherein: step S4 includes the following steps:
aggregating neighbor node information for each graph node by using an attention mechanism according to the time sequence in the sliding window and the current graph structure, and learning out the characteristic representation of the sliding window;
inputting the characteristic representation of the sliding window into the multilayer fully-connected network, and predicting the value of the next time stamp of each graph node.
6. The method of claim 5, wherein: step S5 includes the steps of:
the generator is a graph model based on an attention mechanism, the loss of the generator is calculated, and the generator is updated;
and the GAN model based on the discriminator is used for splicing the real value of the next time stamp after the time sequence in the sliding window to form a real time sequence, splicing the predicted value of the next time stamp after the time sequence in the sliding window to form a predicted time sequence, inputting the real time sequence and the predicted time sequence into the discriminator, calculating the loss of the discriminator and updating the discriminator.
7. The method of claim 6, wherein: step S6 includes the steps of:
performing no masking operation on the time sequence in the sliding window, and calculating the predicted value of each node;
calculating an abnormal score of each node through the difference between the predicted value and the real value;
to reduce the effect of the relative sizes of the different sequences, the present invention uses a smoothing function when computing the anomaly score: subtracting the average value of each sequence and dividing by the standard deviation, and selecting the highest score as a threshold value;
and judging the time point with the abnormal score higher than the threshold value as an abnormal time point to obtain a final abnormal analysis result.
8. A multivariate time series anomaly detection system based on a masked graph neural network model, characterized in that the method of any one of claims 1-7 is performed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211405450.XA CN115935285A (en) | 2022-11-10 | 2022-11-10 | Multi-element time series anomaly detection method and system based on mask map neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211405450.XA CN115935285A (en) | 2022-11-10 | 2022-11-10 | Multi-element time series anomaly detection method and system based on mask map neural network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115935285A true CN115935285A (en) | 2023-04-07 |
Family
ID=86552969
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211405450.XA Pending CN115935285A (en) | 2022-11-10 | 2022-11-10 | Multi-element time series anomaly detection method and system based on mask map neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115935285A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116432870A (en) * | 2023-06-13 | 2023-07-14 | 齐鲁工业大学(山东省科学院) | Urban flow prediction method |
-
2022
- 2022-11-10 CN CN202211405450.XA patent/CN115935285A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116432870A (en) * | 2023-06-13 | 2023-07-14 | 齐鲁工业大学(山东省科学院) | Urban flow prediction method |
| CN116432870B (en) * | 2023-06-13 | 2023-10-10 | 齐鲁工业大学(山东省科学院) | Urban flow prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112527788B (en) | Method and device for detecting and cleaning abnormal value of transformer monitoring data | |
| Chen et al. | A just-in-time-learning-aided canonical correlation analysis method for multimode process monitoring and fault detection | |
| CN111460167A (en) | Method for positioning pollution discharge object based on knowledge graph and related equipment | |
| CN110895526A (en) | Method for correcting data abnormity in atmosphere monitoring system | |
| CN108985380B (en) | Point switch fault identification method based on cluster integration | |
| CN107682319A (en) | A kind of method of data flow anomaly detection and multiple-authentication based on enhanced angle Outlier factor | |
| CN113762329A (en) | Method and system for constructing state prediction model of large rolling mill | |
| Jiang et al. | Electrical-STGCN: An electrical spatio-temporal graph convolutional network for intelligent predictive maintenance | |
| CN105607631B (en) | The weak fault model control limit method for building up of batch process and weak fault monitoring method | |
| CN116823227A (en) | Intelligent equipment management system and method based on Internet of things | |
| CN111638707A (en) | Intermittent process fault monitoring method based on SOM clustering and MPCA | |
| CN116956189A (en) | Current abnormality detection system, method, electronic equipment and medium | |
| CN117421994A (en) | Edge application health monitoring method and system | |
| CN113608968A (en) | Power dispatching monitoring data anomaly detection method based on density and distance comprehensive decision | |
| CN116383645A (en) | Intelligent system health degree monitoring and evaluating method based on anomaly detection | |
| CN116720095A (en) | Electrical characteristic signal clustering method for optimizing fuzzy C-means based on genetic algorithm | |
| CN115935285A (en) | Multi-element time series anomaly detection method and system based on mask map neural network model | |
| CN111984514A (en) | Prophet-bLSTM-DTW-based log anomaly detection method | |
| CN112508278A (en) | Multi-connected system load prediction method based on evidence regression multi-model | |
| CN108123436B (en) | Voltage out-of-limit prediction model based on principal component analysis and multiple regression algorithm | |
| CN117475245A (en) | Intelligent production method and system for preparing flame-retardant polyolefin insulating material | |
| CN106816871B (en) | State similarity analysis method for power system | |
| Xing-yu et al. | Autoencoder-based fault diagnosis for grinding system | |
| CN116956089A (en) | Training method and detection method for temperature anomaly detection model of electrical equipment | |
| KR20230047731A (en) | Apparatus and method for analyzing of load prediction model based on machine learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |