An abnormal power consumption site analysis method and system

By integrating, filling, standardizing, and selecting features from electricity consumption data, and using the ELM algorithm to build a prediction model, the problem of low accuracy and efficiency in identifying abnormal electricity consumption in traditional methods has been solved, thus achieving automation and improved accuracy in electricity consumption inspection.

CN117194971BActive Publication Date: 2025-11-28JINZHOU ELECTRIC POWER SUPPLY COMPANY OF STATE GRID LIAONING ELECTRIC POWER SUPPLY +1
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202310795541.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-06-30
Publication Date
2025-11-28
Estimated Expiration
2043-06-30

AI Technical Summary

Technical Problem

Traditional manual inspections of abnormal electricity usage have low accuracy and efficiency, while neural network models have long parameter adjustment times and weak generalization performance, making it difficult to quickly and accurately identify users with abnormal electricity usage.

Method used

The Extreme Learning Machine (ELM) algorithm is used to integrate, fill, normalize, and select features from electricity consumption data to build a prediction model. The ELM algorithm is then used for training and prediction to determine whether there is abnormal behavior in electricity consumption.

Benefits of technology

It shortened the model training time, improved the accuracy and efficiency of abnormal power consumption identification, and realized the automation and accuracy of power consumption inspection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117194971B_ABST
    Figure CN117194971B_ABST
Patent Text Reader

Abstract

The application discloses a kind of non-normal power consumption field analysis method and system, comprising: obtaining power consumption data;The power consumption data is integrated, and data of different sources is loaded into a new data source;Electricity data filling is carried out in new data source, and abnormal power consumption data is eliminated;The data source is different in the value range of the characteristics of the data that too much difference is normalized;The characteristics of the data after normalization are selected;Selected data is trained using ELM algorithm to obtain a prediction model;Real-time power consumption data is predicted using the obtained prediction model to determine whether there is non-normal power consumption.The non-normal power consumption field analysis method and system based on extreme learning provided by the application not only can reduce the model network parameter adjustment time and enhance the model generalization performance, but also can ensure the accuracy of power consumption non-normal inspection work.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the field of abnormal power consumption detection, in particular to an abnormal power consumption field analysis method and system. BACKGROUND

[0002] The current social power facilities have been very popular, people in daily life, learning and work are inseparable from electricity. The power inspection link is the main component of the entire power supply enterprise, and is also the basis for ensuring normal power consumption of residents. The power consumption enterprise of China plays an important role in the safety guarantee of the power market and the power consumption customers, including the main function of feedback of acceptance. The power supply enterprise has carried out a series of inspection and guidance work to ensure that residents can use electricity normally and safely. The power inspection is also the after-sales service after the power marketing. This service plays an important role in stabilizing customers, improving service quality and promoting marketing. How to quickly and accurately identify the non-normal power consumption users and improve the management efficiency of power inspection is the current problem to be solved.

[0003] For power enterprises, the traditional manual inspection of abnormal power consumption behavior has low accuracy and efficiency. Establishing an intelligent analysis of whether the power consumption is abnormal power consumption inspection system can save human resources, improve the automaticity of information processing, and reduce the operating cost of enterprises.

[0004] Most of the available traditional neural network models adopt gradient descent algorithm, on the basis of analyzing historical power consumption abnormal behavior cases, design power consumption abnormal behavior evaluation index system; use the preprocessed data to construct expert samples; take the abnormal power consumption behavior mark as the input item, take the abnormal power consumption suspicion coefficient as the output item, and construct a neural network model; input the test data into the constructed fuzzy neural network model, and diagnose the abnormal power consumption behavior. The model has the problems of long network parameter adjustment time and weak generalization performance. SUMMARY

[0005] The purpose of the present application is to provide an abnormal power consumption field analysis method and system, which can not only reduce the model network parameter adjustment time and enhance the model generalization performance, but also ensure the accuracy of the abnormal power consumption inspection work.

[0006] In order to solve the above technical problems, the technical scheme adopted by the present application is as follows:

[0007] An abnormal power consumption field analysis method, the specific steps include:

[0008] Obtain power consumption data;

[0009] Integrate the power consumption data, and load the data of different sources into a new data source;

[0010] Fill the power consumption data in the new data source, eliminate abnormal power consumption data;

[0011] Standardize the data with too large value range difference of different features in the data source;

[0012] Select the features of the standardized data;

[0013] Train the selected data using the ELM algorithm to obtain a prediction model;

[0014] Use the obtained prediction model to predict real-time power consumption data to determine whether the power consumption is abnormal.

[0015] Further, the power consumption data includes the electric load value of the user at the corresponding time, the average metered power, and the maximum metered power.

[0016] Further, the power consumption data integration refers to data cleaning and conversion. The cleaning filters the power meter collecting power consumption data and only retains key fields, including user substation number, measurement point identifier, partition unit, data date, metered power, and industry code. The conversion combines two key features: daily power consumption peak and monthly total power consumption according to the original power consumption information. The records with all feature values of 0 in the integrated data table are equivalent to empty records and are deleted together.

[0017] Further, the power consumption data filling includes missing value mean interpolation, abnormal value deletion, filling with uniform default value, and filling with feature statistics.

[0018] Further, during the data standardization process, the value range difference of different features is too large, which means that the power consumption of large enterprise customers and the power consumption of residential customers differ by more than 10,000 degrees to prevent the features with large values from being lost in features with small values.

[0019] Further, during the data standardization process, the normalized value = (original value - minimum value) / (maximum value - minimum value) is used to normalize the load value to the [0, 1] interval. If the value of a certain feature is all 0, it will cause the denominator of the normalization formula to be 0, at which time the feature is not normalized and the original value is kept as 0.

[0020] Further, the feature selection of the data is based on the feature correlation method.

[0021] Further, the prediction model is:

[0022]

[0023] where: w = (w ij ) L*nw represents the weight between the hidden layer and the input neuron i = [w i1 ,w i2 ,…,w in ] T b represents the input weight between the i-th node of the hidden layer and the input neuron, b = [b1, b2, …, b L ] T b represents the bias value, f(x) represents the activation function, w i x j represents the inner product between the two.

[0024] An abnormal power consumption site analysis system of the abnormal power consumption site analysis method, comprising:

[0025] An electricity data acquisition module for acquiring electricity data;

[0026] An electricity data integration module for integrating the acquired electricity data, loading data from different data sources into a new data source;

[0027] An electricity data filling module for eliminating abnormal electricity data;

[0028] A feature normalization module for normalizing data with excessively large value range differences of different features;

[0029] A feature selection module for selecting features of the normalized data;

[0030] A prediction model generation module for training the processed data using an ELM algorithm to obtain a prediction model;

[0031] An abnormal electricity prediction module for predicting other electricity data using the obtained prediction model to determine whether the electricity is abnormal.

[0032] Further, the electricity data integration module comprises:

[0033] A data cleaning module, which filters the electricity meter collecting electricity data and only retains key fields, including user substation number, measurement point identification, partition unit, data date, copied electricity quantity, and industry code;

[0034] A data conversion module, which combines two key features: daily electricity peak and monthly electricity total according to the original electricity quantity information, and deletes all records with feature values of 0 in the integrated data table.

[0035] The application has the following beneficial effects: the abnormal power consumption site analysis plays a very important role in the field of electric power, which not only can reduce the model network parameter adjustment time and enhance the model generalization performance, but also can ensure the accuracy of the abnormal power consumption inspection work. In order to improve the efficiency and accuracy of the abnormal power consumption site analysis, the application provides an abnormal power consumption site analysis method based on extreme learning. First, the input original power consumption data is processed, including data integration, data filling, feature normalization and feature selection, and then the data is divided into a training set, a test set and a validation set; secondly, the processed data is trained by using the ELM algorithm to obtain a prediction model; finally, the abnormal user information and the normal user information predicted by the algorithm are compared to obtain a conclusion. The experimental results show that the ELM algorithm has high reliability. The application can analyze and arrange a large amount of power consumption data, find the main features of abnormal power consumption, realize automatic abnormal inspection of power consumption site, improve the accuracy of abnormal power consumption identification, reduce the time required for training the model, facilitate the staff, improve the business efficiency of the enterprise, and improve the image of the enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A flowchart of an abnormal power consumption site analysis method provided by the application;

[0037] Figure 2 An architecture diagram of an abnormal power consumption site analysis system provided by the application. DETAILED DESCRIPTION

[0038] The principles and characteristics of the application are described below in combination with the drawings, and the examples are only used to explain the application and not to limit the scope of the application.

[0039] The application provides an abnormal power consumption detection field, and an abnormal power consumption site analysis method based on extreme learning, as shown in Figure 1 , which comprises:

[0040] S1: obtaining power consumption data;

[0041] The power consumption data comprises the electric load value of the user, the average copy seen power, and the maximum value of the copy seen power. The power consumption curves of normal users are generally similar, the electric load value of the abnormal power consumption behavior detection is larger than that of the normal user, and the fluctuation is also larger.

[0042] S2: power consumption data integration;

[0043] In the embodiment, the data integration is to load the data of different sources (factory, merchant and household electricity) into a new data source after cleaning and conversion. The cleaning is to filter the electricity meter collecting electricity data and only keep the key fields, including user substation number, measurement point identification, partition unit, data date, copied power and industry code. The conversion is to combine two key features according to the original electricity quantity information: daily electricity peak and monthly electricity total. The record with all feature values of 0 in the integrated data table is equivalent to an empty record and is deleted together.

[0044] S3: Electricity data filling;

[0045] The electricity data filling in the data source includes: missing value mean interpolation, abnormal value deletion, filling with a unified default value and filling with a statistical quantity of a feature, so as to eliminate abnormal electricity data, which refers to data missing or missing, to ensure that the subsequent algorithm can be normally analyzed.

[0046] S4: The data with too large value range difference of different features in the data source is normalized, the too large value range difference of different features refers to the electricity quantity of large enterprise customers and the electricity quantity of residential customers being more than 10000 degrees, to prevent the feature with large value from hiding the feature with small value. According to the normalization value=(original value-minimum value) / (maximum value-minimum value), the load value is normalized to the interval [0,1]; if the value of a certain feature is all 0, it will cause the denominator of the normalization formula to be 0, at this time, the feature is not normalized, and the original value is kept as 0.

[0047] S5: The features of the normalized data are selected by using a method based on feature correlation

[0048] When the number of original data features is large, the correlation between the features will bring difficulties to the analysis, and the running time of the algorithm will also be significantly increased, so it is necessary to select the features in the original data, that is, feature selection based on feature correlation. Since the method based on feature correlation is commonly used and has good effect, the method based on feature correlation is used for feature selection.

[0049] Feature selection based on feature correlation is a classical filter mode feature selection algorithm, which evaluates the effect of a single feature on each classification to obtain the final feature subset, and the formal evaluation method is as follows:

[0050]

[0051] Merit (S) = ∑i=1kfi(S) where f (S) is the evaluation of a feature subset S containing k features. s Merit (S) = ∑i=1kfi(S) where f (S) is the evaluation of a feature subset S containing k features. denotes the feature-subset average correlation corresponding to the subset, where f ∈ S; denotes the average feature-feature correlation. All variables given by the above formula are standardized. This evaluation index can effectively represent the contribution value of the given feature to classification and eliminate irrelevant or features with small contribution values; all features in the above formula must be discrete random variables. If they are numerical variables, they must be discretized first, and the correlation between features is evaluated by the entropy calculation method.

[0052] S6: Use the ELM algorithm to train the data after feature selection to obtain a prediction model:

[0053] Compared with traditional neural networks, especially the single-hidden-layer feedforward neural network SLFNs, the ELM algorithm has the advantages of fast speed and better generalization performance under the condition of ensuring accuracy.

[0054] For a single-hidden-layer neural network, assume there are N arbitrary samples (x i , t i ), x i = [x i1 , x i2 , …, x in T ∈R n , which means that an SLFNs with L (L < N) hidden layer nodes can be represented in model form as:

[0055]

[0056] where, w = (w ij ) L*n denotes the weight between the hidden layer and the input neurons, w i = [w i1 , w i2 , …, w in T is the input weight connecting the i-th node in the hidden layer and the input neurons, b = [b1, b2, …, b L T is the bias value, f(x) is the activation function, w i x j denotes the inner product between the two.

[0057] However, the purpose of the hidden layer neural network is to minimize the output result error, that is:

[0058]

[0059] There exists β i , w i , b i to make​​​

[0060]

[0061] In matrix form, we have:

[0062] Hβ = T

[0063] where H is the output of the hidden layer:

[0064]

[0065] The output weights β = (β1, …, β L ) T , T is the desired output, T = (T1, …, T N ) T .

[0066] Traditional neural networks are almost all based on gradient descent, but gradient descent requires adjusting all parameters in the iterative process, while in the ELM algorithm, the weights and the bias values of the hidden layer are randomly determined, and then it becomes a linear system formula:

[0067]

[0068] Thus, the output weights can be solved, β = H + T, H + is the Moore-Penrose generalized inverse of the matrix H.

[0069] The algorithm of ELM is as follows:

[0070]

[0071] S7: use the model to make prediction:

[0072] After training, a model for predicting power consumption data is obtained, and the model is used to predict the data to determine whether the user is using electricity abnormally.

[0073] As shown in Figure 2 , a non-normal power consumption field analysis system of the non-normal power consumption field analysis method as described above, comprising:

[0074] A power consumption data acquisition module for acquiring power consumption data;

[0075] A power consumption data integration module for integrating the acquired power consumption data, loading data from different data sources into a new data source;

[0076] A power consumption data filling module for eliminating abnormal power consumption data;

[0077] The feature normalization module normalizes data with too large value range difference of different features;

[0078] The feature selection module selects features of the normalized data;

[0079] The prediction model generation module trains the processed data using an ELM algorithm to obtain a prediction model;

[0080] The electricity abnormality prediction module uses the obtained prediction model to predict other electricity data to determine whether electricity is abnormal.

[0081] Further, the electricity data integration module comprises:

[0082] The data cleaning module filters the electricity meter collecting electricity data and only retains key fields, including user substation number, measurement point identifier, partition unit, data date, copied seen electricity and industry code;

[0083] The data conversion module combines two key features, daily electricity peak and monthly total electricity consumption, according to the original electricity consumption information, and deletes records with all feature values of 0 in the integrated data table.

[0084] In order to verify the effectiveness of the present application, the present application uses the data of a certain power enterprise for experiment, and the experiment proves that the difference between the electricity consumption of the abnormal user predicted by the ELM algorithm provided by the present application and the electricity consumption of the normal user is more significant, and the electricity consumption of the abnormal user predicted by the algorithm is more close to the electricity consumption of the real abnormal user, so that the credibility of the present application is higher.

[0085] The experimental data comes from a certain power enterprise, and we have preprocessed the data as input data of the algorithm, and each record has 26 fields, including user number, average copied seen electricity in a year and maximum copied seen electricity. The data with three commercial electricity price codes 200, 300 and 370 are merged with the real abnormal user data to form a training set; the three commercial electricity price code data correspond to three training sets, and the data set with the electricity price code 001 is used as the test set after the model is trained.

[0086] Since the low-voltage user power consumption statistical table is large, and many fields are not needed in power consumption analysis, these tables are filtered, only keeping the key fields such as user substation number, measurement point identification, partition unit, data date, copied apparent power and industry code. In addition, since the original data is daily statistical power consumption, all as features will cause feature inflation, affecting the quality of data mining, we combine two key features according to the original power consumption information: daily power consumption peak and monthly total power consumption. The record with all feature values of 0 in the integrated data table is equivalent to an empty record, which has little effect on analysis and is also deleted.

[0087] For missing values, fill in 0; for abnormal values, since the records containing abnormal values are few, the records containing abnormal values are directly deleted.

[0088] According to the standard value=(original value-minimum value) / (maximum value-minimum value), the load value is normalized to the interval [0,1]. If the value of a certain feature is all 0, it will cause the denominator of the normalization formula to be 0, at which time the feature is not normalized, that is, the original value is kept as 0.

[0089] Since the data of real abnormal power users is obtained by on-site investigation of power law enforcement personnel, it is limited by manpower, and the investigation personnel can only investigate a very small part of suspicious personnel. Therefore, the real abnormal user data set that can be compared is only a very small subset of the actual real abnormal user data set. In order to verify the analysis conclusion, the abnormal user information predicted by the algorithm is compared with the normal user information, and the comparison between the monthly average power consumption and the monthly maximum power consumption of the user is obtained. It is found that the difference between the power consumption of the abnormal user predicted by the ELM algorithm provided by the application and the power consumption of the normal user is more significant, and the power consumption of the abnormal user predicted by the algorithm is more close to the power consumption of the real abnormal user, so it is concluded that the credibility of the application is high.

[0090] Those skilled in the art should understand that the embodiments of the application can be provided as a method, a system, or a computer program product. Therefore, the application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code. The solutions in the embodiments of the application can be implemented in various computer languages, such as object-oriented programming language Java and interpreted scripting language JavaScript.

[0091] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0092] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0093] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flowchart illustrations and / or block diagrams block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0094] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those of skill in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, the attached claims are intended to embrace all such variations and modifications as fall within the scope of the application.

[0095] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A method of analyzing a site of illegal electricity use, characterized by, The method comprises the following steps: acquiring power consumption data, the power consumption data including the electric load value of a user at a corresponding time, the average meter reading power, and the maximum meter reading power; integrating the power consumption data to load data from different sources into a new data source; the power consumption data integration refers to data cleaning and conversion, the cleaning filters the electric meter collecting power consumption data and only retains key fields, the key fields including user substation number, measurement point identification, partition unit, data date, meter reading power, and industry code, the conversion combines two key features: daily power consumption peak and monthly power consumption total according to original power consumption information, and all feature values of 0 in the integrated data table are equivalent to empty records and are deleted together; performing power consumption data filling in the new data source to eliminate abnormal power consumption data; the power consumption data filling includes missing value mean interpolation, abnormal value deletion, filling with a unified default value, and filling with a characteristic statistical quantity; performing normalization processing on data with a too large value range difference of different features in the data source; the too large value range difference of different features refers to a difference of more than 10,000 degrees between the power consumption of a large enterprise customer and the power consumption of a residential customer to prevent a large value feature from overshadowing a small value feature; during normalization processing, the load value is normalized to the [0, 1] interval according to the normalized value=(original value-minimum value) / (maximum value-minimum value); if the value of a certain feature is all 0, the denominator of the normalization formula will be 0, and the feature is not normalized at this time, and the original value is kept as 0; using selecting features of the normalized data using a feature correlation-based method; training the selected data using an ELM algorithm to obtain a prediction model: where: denotes the weights between the hidden layer and the input neurons, is the input weight connecting the i-th node of the hidden layer and an input neuron, is the bias value, is the activation function, denotes the inner product between the two; using the obtained prediction model to predict real-time power consumption data to determine whether there is a power consumption violation.

Citation Information

Patent Citations

  • Electric quantity abnormity detection method based on Ensemble learning model

    CN110119755A