Anti-electricity-stealing analysis method and system based on big electricity data and intelligent algorithm model

Abstract

This invention discloses a method and system for anti-electricity theft analysis based on big data of electricity consumption and intelligent algorithm models. The method includes: analyzing line loss data within a region to obtain a first electricity theft analysis result; acquiring user electricity consumption data and accompanying factor data, wherein the electricity consumption data includes time-series data of electricity consumption; analyzing the user electricity consumption data and accompanying factor data using an intelligent algorithm model to obtain a second electricity theft analysis result; determining the existence of electricity theft based on the first and second electricity theft analysis results; and, if electricity theft is determined, analyzing and locating the location and time period of electricity theft based on line loss data, user electricity consumption data, and accompanying factor data within the region. This invention, by first analyzing line loss within the transformer area and then analyzing user data within the transformer area, effectively improves the efficiency of electricity theft analysis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electricity stealing analysis, and particularly relates to an anti-electricity stealing analysis method and system based on electricity big data and intelligent algorithm model. BACKGROUND

[0002] Electricity stealing behavior is an important part of non-technical loss of the power system, and a series of electricity stealing behaviors on the distribution network side will not only cause huge economic losses, but also bring immeasurable safety hazards. Although the existing technology has made certain progress in electricity stealing detection, there are still some deficiencies: first, the traditional detection methods such as line tapping or manual inspection of electricity meter tampering are low in efficiency and high in cost. Second, with the popularization of smart grids, the amount of user electricity data has increased sharply, and the huge amount of data has increased the difficulty of data processing and model establishment; third, the types of electricity stealing behaviors are various, and the conventional two-class electricity stealing model often cannot accurately capture the subtle differences between different electricity stealing behaviors, making it difficult to meet the needs of power companies. SUMMARY

[0003] In view of the problems existing in the prior art, the present application provides an anti-electricity stealing analysis method and system based on electricity big data and intelligent algorithm model, which effectively improves the anti-electricity stealing analysis efficiency and accuracy. The technical scheme is as follows:

[0004] In the first aspect, an anti-electricity stealing analysis method based on electricity big data and intelligent algorithm model is provided, comprising the following steps:

[0005] Analyzing a first electricity stealing analysis result for line loss data in the region;

[0006] Obtaining electricity data and accompanying factor data of the user, wherein the electricity data includes electricity consumption time series data; and the accompanying factor data includes meteorological, calendar and household attribute data accompanying the user's electricity consumption at the same time;

[0007] Analyzing a second electricity stealing analysis result based on the electricity data and accompanying factor data of the user by using an intelligent algorithm model;

[0008] Determining that there is electricity stealing based on the first electricity stealing analysis result and the second electricity stealing analysis result;

[0009] When it is determined that there is electricity stealing, analyzing and positioning the electricity stealing location and time period based on the line loss data in the region, the electricity data and the accompanying factor data of the user.

[0010] In some embodiments, the analyzing a second electricity stealing analysis result based on the electricity data and accompanying factor data of the user by using an intelligent algorithm model comprises:

[0011] Based on the daily, weekly, and monthly electricity consumption time-series data of all users, a trend similarity analysis is performed. Users with similar trends are grouped into one category. Specifically, if the trends of a user's daily, weekly, and monthly electricity consumption time-series data are all similar, the user is grouped into one category. If the trend of any one of the three is not similar, the user is not grouped into one category and is identified as a candidate electricity theft user.

[0012] A separate class is formed for users who are not assigned to any class. Users in a class with fewer than a first preset value are identified as candidate electricity theft users.

[0013] For candidate electricity theft users, obtain the user's historical electricity consumption time-series data and historical accompanying factor time-series data;

[0014] Based on users' historical electricity consumption time-series data and historical accompanying factor time-series data, we can mine the electricity consumption patterns of users under the influence of accompanying factors.

[0015] Based on the time-series data of electricity consumption and accompanying factors of candidate electricity theft users in the current time period, the user patterns are used to analyze whether candidate electricity theft users have committed electricity theft in the current time period.

[0016] In some implementations, the trend similarity analysis includes:

[0017] Calculate the growth rate of the daily electricity consumption time series data of the first user from the previous time to the next time, and obtain the daily electricity consumption time series growth rate value sequence of the first user; calculate the growth rate of the daily electricity consumption time series data of the second user from the previous time to the next time, and obtain the daily electricity consumption time series growth rate value sequence of the second user.

[0018] Based on the time-series growth rate values ​​of the first user's daily electricity consumption and the second user's daily electricity consumption, the difference between the time-series growth rate values ​​of the first user's daily electricity consumption and the second user's daily electricity consumption at the same time is calculated. The sum of the differences corresponding to all times in the daily electricity consumption time-series data is used as the similarity between the trend of the first user's daily electricity consumption time-series data and the trend of the second user's daily electricity consumption time-series data.

[0019] In some implementations, the method of mining user electricity consumption patterns under the influence of accompanying factors based on historical electricity consumption time-series data and historical accompanying factor time-series data includes:

[0020] Based on the historical electricity consumption time series data and historical accompanying factor time series data of different users as the first dataset, the electricity consumption pattern of users under the influence of accompanying factors is mined through neural network model training, and recorded as the first electricity consumption pattern. The first electricity consumption analysis model representing the first electricity consumption pattern is obtained.

[0021] The first electricity consumption analysis model, which represents the first electricity consumption pattern, is used as the base model. The second dataset is based on the historical electricity consumption time series data and historical accompanying factor time series data of the same user. The base model is transformed to explore the electricity consumption pattern of the same user under the influence of accompanying factors, which is denoted as the second electricity consumption pattern. The second electricity consumption analysis model representing the second electricity consumption pattern is then obtained.

[0022] In some implementations, the basic model is transformed to obtain a second electricity consumption analysis model characterizing the second electricity consumption pattern, including:

[0023] Based on the second sample input of the second dataset, the final encoding result of the basic model encoding the electricity consumption time series data and the accompanying factor time series data is obtained, which is the result before the output layer of the basic model.

[0024] The final encoding result based on the basic model is transformed by the first linear transformation matrix to output the first encoding transformation result;

[0025] Based on the first encoding transformation result, a nonlinear transformation is performed through a nonlinear activation function layer to obtain the nonlinear encoding result;

[0026] Based on the nonlinear encoding result, a first linear inverse transformation matrix opposite to the first linear transformation matrix is ​​used to obtain the first decoding result. The final encoding result of the basic model and the first decoding result are combined as the fused first decoding result. The fused first decoding result is input into the output layer of the basic model for fully connected processing to obtain the electricity theft identification result of the second sample.

[0027] The second electricity consumption analysis model is obtained by training the first linear transformation matrix, the nonlinear activation function layer, and the first linear inverse transformation matrix based on the error of the second sample electricity theft labeling data and the actual electricity theft identification results.

[0028] In some implementations, the step of creating a separate class based on users not assigned to any class, and identifying users in a class with fewer than a first preset number as candidate electricity theft users, further includes:

[0029] Based on users in a class whose number of users is greater than a first preset value, determine the height distance between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curve of the other users in the class, and obtain the sum of the height distances between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curves of all other users in the class;

[0030] The sum of the height distances corresponding to the electricity consumption time-series data curves of each user are sorted from largest to smallest, and the electricity consumption time-series data curves of users whose values ​​are at the top of the sorted preset values ​​are selected as candidate electricity theft users.

[0031] In some implementations, the step of mining the user's electricity consumption patterns under the influence of accompanying factors based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data includes: using the user's historical electricity consumption time-series data and historical accompanying factor time-series data as input to train a neural network-based intelligent algorithm model to mine the user's electricity consumption patterns, and characterizing the user's electricity consumption patterns based on the trained neural network model.

[0032] The neural network-based intelligent algorithm includes:

[0033] The first input data is the electricity consumption time series data and the accompanying factor time series data over a preset long period.

[0034] Obtain the increase value of the electricity consumption time series data and the accompanying factor time series data of the next moment compared to the previous moment within a preset long time period; obtain the electricity consumption time series growth rate data and the accompanying factor growth rate data of the preset long time period; and use the electricity consumption time series growth rate data and the accompanying factor growth rate data as the second input data.

[0035] Each time series data of the first input data is input into the sequence feature extraction unit within a preset long time period to extract the time series features of each time series data, thereby characterizing the time relationship features of each time series data.

[0036] The electricity consumption time-series growth rate data and the accompanying factor growth rate data at the same time as the second input data are input into the convolutional layer unit to obtain the spatial relationship features of the electricity consumption time-series growth rate data and the accompanying factor growth rate data.

[0037] The temporal and spatial relationship features of the input data are fed into a fully connected layer for analysis and judgment to determine whether the input data is characteristic data of electricity theft.

[0038] Secondly, an anti-electricity theft analysis system based on big data on electricity consumption and intelligent algorithm models is provided. This system includes:

[0039] The first electricity theft analysis unit is used to analyze the line loss data within the area to obtain the first electricity theft analysis result;

[0040] The electricity consumption data acquisition unit is used to acquire users' electricity consumption data and accompanying factor data. The electricity consumption data includes electricity consumption time-series data; the accompanying factor data includes weather, calendar, and household attribute data that accompany the user's electricity consumption at the same time.

[0041] The second electricity theft analysis unit is used to analyze the second electricity theft analysis results based on the user's electricity consumption data and accompanying factor data using an intelligent algorithm model.

[0042] The third electricity theft analysis unit is used to determine the existence of electricity theft based on the results of the first and second electricity theft analyses.

[0043] The electricity theft location unit is used to analyze and locate the location and time period of electricity theft based on line loss data, user electricity consumption data and accompanying factor data within the area when electricity theft is confirmed.

[0044] Thirdly, an electronic device is provided, the electronic device comprising:

[0045] processor;

[0046] Memory used to store processor-executable instructions;

[0047] The processor executes the executable instructions to implement the anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm models as described in the first aspect above.

[0048] Fourthly, a computer-readable storage medium is provided, on which computer instructions are stored, which, when executed by a processor, implement the steps of the anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm model as described in the first aspect above.

[0049] This invention discloses an anti-electricity theft analysis method and system based on big data on electricity consumption and intelligent algorithm models, which has the following beneficial effects: First, based on the line loss calculation results of the transformer area, this invention determines whether there are any abnormal line losses. If abnormal line losses are found, further analysis is performed targeting users in the transformer area. Considering factors such as weather, temperature, calendar, time, and quarterly factors, the analysis examines whether the users' electricity consumption conforms to normal electricity consumption patterns. This invention's approach of first analyzing the line loss of the transformer area and then analyzing the users effectively improves the efficiency of electricity theft analysis. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm models in the embodiments of this application;

[0051] Figure 2 This is a flowchart illustrating the method for analyzing the second electricity theft analysis result using an intelligent algorithm model based on the user's electricity consumption data and accompanying factor data in this application embodiment.

[0052] Figure 3 This is a flowchart illustrating the time-series data trend similarity analysis method in an embodiment of this application;

[0053] Figure 4 This is a schematic diagram of the method for mining user electricity consumption patterns under the influence of accompanying factors based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data in this application embodiment;

[0054] Figure 5This is a schematic diagram of the anti-electricity theft analysis system based on big data on electricity consumption and intelligent algorithm model in the embodiments of this application. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0056] This application provides an anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm models, which includes the following steps:

[0057] Step 1: Analyze the line loss data within the area to obtain the first electricity theft analysis results;

[0058] Step 2: Obtain the user's electricity consumption data and accompanying factor data. The electricity consumption data includes time-series electricity consumption data. The accompanying factor data includes weather, calendar, and family attribute data (such as the composition of family members, income, etc.) that accompany the user's electricity consumption at the same time.

[0059] Step 3: Based on the user's electricity consumption data and accompanying factor data, analyze the results of the second electricity theft analysis using an intelligent algorithm model;

[0060] Step 4: Based on the results of the first and second electricity theft analyses, determine that electricity theft has occurred;

[0061] Step 5: If electricity theft is confirmed, analyze and locate the location and time period of electricity theft based on line loss data, user electricity consumption data and accompanying factor data in the area.

[0062] In this embodiment, the existence of line loss anomalies is first determined based on the calculation results of the transformer substation line loss. If anomalies exist, considering that electricity theft would lead to an increase in grid line loss, this application further analyzes the transformer substation users as the target objects, considering factors such as weather, temperature, calendar, time, and quarter to analyze whether the users' electricity consumption conforms to normal electricity consumption patterns. This approach of first analyzing the transformer substation line loss and then analyzing the users in the substation effectively improves the efficiency of electricity theft analysis.

[0063] In one implementation, step 3 above, based on the user's electricity consumption data and accompanying factor data, analyzes the second electricity theft analysis result using an intelligent algorithm model, including the following steps:

[0064] Step 31: Perform trend similarity analysis based on the daily, weekly, and monthly electricity consumption time series data of all users, and classify users with similar trends into one category. Specifically, if the daily, weekly, and monthly electricity consumption time series data trends of a user are all similar, then the user is classified into one category. If the trend of any one of the three is not similar, then the user is not classified into one category and is identified as a candidate electricity theft user.

[0065] Step 32: Create a separate class based on users who are not assigned to a class. For users in a class with fewer than the first preset value, identify them as candidate electricity theft users.

[0066] Step 33: For candidate electricity theft users, obtain the user's historical electricity consumption time-series data and historical accompanying factor time-series data;

[0067] Step 34: Based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data, mine the user's electricity consumption patterns under the influence of accompanying factors.

[0068] Step 35: Based on the time-series data of electricity consumption and accompanying factors of the candidate electricity theft users in the current time period, and using the user patterns, analyze whether the candidate electricity theft users have committed electricity theft in the current time period.

[0069] It's understandable that different users have different electricity consumption patterns. For example, some users consume more electricity on weekends, office workers consume less during the day and more in the morning and evening, while commercial users generally consume a relatively consistent amount each day. The load curve fluctuations of electricity thieves and legitimate users are significantly different; legitimate users exhibit distinct characteristic fluctuation patterns. Based on the similarity analysis of different users' electricity consumption trends, those grouped into a single category represent users within that category whose consumption patterns are similar. Users not grouped into a single category, or whose categories contain too few users, and whose consumption patterns do not conform to the majority of users' patterns, are likely electricity thieves.

[0070] It is understandable that the daily, weekly, and monthly electricity consumption time-series data trends of normal electricity users will be similar to those of other users with similar electricity consumption patterns. However, for electricity thieves, one of their daily, weekly, and monthly electricity consumption time-series data trends may be disguised as similar to those of other users with similar electricity consumption patterns. However, the electricity consumption change characteristics extracted from multiple observation dimensions will still show anomalies compared to the electricity consumption change characteristics of normal users.

[0071] In one implementation, the trend similarity analysis in step 31 above includes the following steps:

[0072] Step 311: Calculate the growth rate of the next time step in the daily electricity consumption time series data of the first user compared to the previous time step, and obtain the daily electricity consumption time series growth rate value sequence of the first user; calculate the growth rate of the next time step in the daily electricity consumption time series data of the second user compared to the previous time step, and obtain the daily electricity consumption time series growth rate value sequence of the second user.

[0073] Step 312: Based on the daily electricity consumption time-series growth rate value sequence of the first user and the daily electricity consumption time-series growth rate value sequence of the second user, calculate the difference between the daily electricity consumption time-series growth rate value of the first user and the daily electricity consumption time-series growth rate value of the second user at the same time. Obtain the sum of the differences corresponding to all times in the daily electricity consumption time-series data as the similarity between the daily electricity consumption time-series data trend of the first user and the daily electricity consumption time-series data trend of the second user.

[0074] The method for calculating the trend similarity of time-series data in the embodiments of this application includes:

[0075] Calculate the growth rate of the daily electricity consumption time series data of the first user from the previous time to the next time, and obtain the daily electricity consumption time series growth rate value sequence of the first user; calculate the growth rate of the daily electricity consumption time series data of the second user from the previous time to the next time, and obtain the daily electricity consumption time series growth rate value sequence of the second user.

[0076] Based on the time-series growth rate values ​​of the first user's daily electricity consumption and the second user's daily electricity consumption, the difference between the time-series growth rate values ​​of the first user's daily electricity consumption and the second user's daily electricity consumption at the same time is calculated. The sum of the differences corresponding to all times in the daily electricity consumption time-series data is used as the similarity between the trend of the first user's daily electricity consumption time-series data and the trend of the second user's daily electricity consumption time-series data.

[0077] Similarly, calculate the similarity between the weekly electricity consumption time-series data trends of the first user and the second user.

[0078] It is understood that in this embodiment, considering the similarity of the trends of two daily electricity consumption time series data, correlation calculation methods such as Euclidean distance or Pearson coefficient are not used for analysis. Euclidean distance measures the difference in the numerical distribution of two daily electricity consumption time series data, but cannot analyze whether the changing trends of two daily electricity consumption time series data are similar. Pearson coefficient analyzes the correlation of changes in two daily electricity consumption time series data, but cannot accurately determine the degree of similarity or difference in the trends of two daily electricity consumption time series data. In this embodiment, the difference in the growth rate values ​​of two daily electricity consumption time series data curves at the same time is used to measure the degree of difference in the trends of two daily electricity consumption time series data. It is understood that in this embodiment, if at the same time, the growth rate value on one daily electricity consumption time series data curve is greater than 0 (e.g., 4), and the growth rate value on the other daily electricity consumption time series data curve is less than 0 (e.g., -3), it indicates that the trends of the two daily electricity consumption time series data at that time are very different (trend difference is 7), one is an increasing trend, and the other is a decreasing trend. The difference in the growth rate values ​​of the two daily electricity consumption time series data curves at the same time can be used to measure the degree of difference in the trends of two daily electricity consumption time series data.

[0079] In one implementation, step 34 above, based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data, mines the user's electricity consumption patterns under the influence of accompanying factors, including the following steps:

[0080] Step 3401: Based on the historical electricity consumption time series data and historical accompanying factor time series data of different users as the first dataset, the user's electricity consumption pattern under the influence of accompanying factors is mined through neural network model training, and recorded as the first electricity consumption pattern. The first electricity consumption analysis model representing the first electricity consumption pattern is obtained.

[0081] Step 3402: Using the first electricity consumption analysis model representing the first electricity consumption pattern as the base model, and using the historical electricity consumption time series data and historical accompanying factor time series data of the same user as the second dataset, the base model is transformed to mine the electricity consumption pattern of the same user under the influence of accompanying factors, which is recorded as the second electricity consumption pattern, and the second electricity consumption analysis model representing the second electricity consumption pattern is obtained.

[0082] In this embodiment, user electricity consumption patterns are mined from two datasets. The first dataset consists of historical electricity consumption time-series data and historical accompanying factor time-series data of different users, while the second dataset consists of historical electricity consumption time-series data and historical accompanying factor time-series data of the same user. It can be understood that the first electricity consumption rule mined based on the first dataset represents the user's electricity consumption when the accompanying factor values ​​are different. The first dataset is used to mine the correlation between changes in different accompanying factors and the user's electricity consumption. However, the first electricity consumption pattern mined based on the first dataset cannot be matched with the electricity consumption habits of individual users in a refined manner.

[0083] In this application, a second dataset is used to mine a second electricity consumption pattern that is refined to fit the electricity consumption habits of individual users. In this second dataset, some accompanying factors will remain basically unchanged for a long period of time (e.g., family composition, income and other family attribute data), and then change at some time, causing changes in the user's electricity consumption behavior. The second dataset has insufficient diversity of sample data, making it difficult to fully mine the user's second electricity consumption pattern.

[0084] In this embodiment of the application, the first electricity consumption rule is continuously optimized based on the first electricity consumption rule that conforms to different users, and the second electricity consumption rule mined from the second dataset with a small number of samples is used as a guide to form the second electricity consumption rule that conforms to the data characteristics of the second dataset.

[0085] In one implementation, the guiding process may be as follows: training a neural network intelligent algorithm based on a first dataset to obtain a first intelligent analysis network model for predicting user electricity consumption under the influence of various related factors; based on the first intelligent analysis network model, fine-tuning the model parameters of the original first intelligent analysis network model with a second dataset containing less sample data to obtain an optimized second intelligent analysis network model that conforms to the data characteristics of the second dataset, wherein the second intelligent analysis network model is used to characterize the second electricity consumption pattern.

[0086] In one implementation, step 3402 above involves transforming the basic model to obtain a second electricity consumption analysis model characterizing the second electricity consumption pattern, including:

[0087] Step 34021: Input the second sample from the second dataset (the sample in the second dataset is denoted as the second sample) into the basic model to obtain the final encoding result of the basic model encoding the electricity consumption time series data and the accompanying factor time series data, which is the result before the output layer of the basic model;

[0088] Step 34022: Based on the final encoding result of the basic model, the first encoding transformation result is output after passing through the first linear transformation matrix;

[0089] Step 34023: Based on the first encoding transformation result, perform a nonlinear transformation through a nonlinear activation function layer to obtain the nonlinear encoding result;

[0090] Step 34023: Based on the nonlinear encoding result, a first linear inverse transformation matrix opposite to the first linear transformation matrix is ​​used to obtain the first decoding result. The final encoding result of the basic model and the first decoding result are combined as the fused first decoding result. The fused first decoding result is input into the output layer of the basic model for fully connected processing and the electricity theft identification result of the second sample is obtained.

[0091] Step 34024: Based on the error between the electricity theft labeling data of the second sample and the actual electricity theft identification results, train the first linear transformation matrix, the nonlinear activation function layer, and the first linear inverse transformation matrix to obtain the second electricity consumption analysis model.

[0092] In this embodiment, a first electricity consumption analysis model characterizing a first electricity consumption pattern is used as the base model. When mining a second electricity consumption pattern to obtain a second electricity consumption analysis model, the model parameters of the base model remain unchanged. Based on the base model, the existing first electricity consumption analysis model is locally improved. In this embodiment, the encoding result of the base model for the input data is further subjected to linear transformation, nonlinear transformation, and inverse linear transformation. Combining linear and nonlinear transformations improves the ability to express the characteristic differences between the historical electricity consumption time series data and historical accompanying factor time series data of the same user and the historical electricity consumption time series data and historical accompanying factor time series data of different users. Furthermore, in this embodiment, only the parameters of the first linear transformation matrix and the first inverse linear transformation matrix need to be obtained. The final encoding result of the base model and the first decoding result are combined as the fused first decoding result, ensuring that the data input to the output layer of the base model is a fine-tuning of the final encoding result of the base model.

[0093] In one implementation, step 32 above, which involves creating a separate class based on users not assigned to any class, and identifying users in a class with fewer than a first preset value as candidate electricity theft users, further includes:

[0094] Step 321: Based on users in a class with more than one preset value, determine the height distance between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curve of the other users in the class, and obtain the sum of the height distances between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curves of all other users in the class.

[0095] Step 322: Sort the sum of the height distances corresponding to the electricity consumption time-series data curves of each user from largest to smallest, and obtain the electricity consumption time-series data curves of users whose scores are at the top of the preset values ​​as candidate electricity theft users.

[0096] In this embodiment, we consider users within a class whose electricity consumption time-series data show similar trends. However, some users steal electricity by reducing their actual consumption by a fixed percentage based on normal consumption behavior. In this case, the trend of the electricity consumption time-series curve for the electricity thief is the same as that of the normal user, differing only in the magnitude of the electricity consumption. In this embodiment, we determine whether the distribution position of the user's electricity consumption time-series data curve is within a densely distributed area of ​​all user electricity consumption time-series data curves by calculating the sum of the height distances corresponding to each user's electricity consumption time-series data curve. If the sum of the height distances corresponding to a user's electricity consumption time-series data curve is smaller, then the distribution position of that user's electricity consumption time-series data curve is within a densely distributed area of ​​all user electricity consumption time-series data curves. The height distance is determined based on the difference in electricity consumption values ​​at the same moment on two user electricity consumption time-series data curves, and further, the sum of the differences at all moments on the electricity consumption time-series data curves is used as the height distance between the two user electricity consumption time-series data curves.

[0097] In one implementation, step 34 above, which mines the user's electricity consumption patterns under the influence of accompanying factors based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data, includes: using the user's historical electricity consumption time-series data and historical accompanying factor time-series data as input, training a neural network-based intelligent algorithm model to mine the user's electricity consumption patterns, and characterizing the user's electricity consumption patterns based on the trained neural network model.

[0098] The neural network-based intelligent algorithm includes:

[0099] Step 3411: Use electricity consumption time-series data and accompanying factor time-series data over a preset long period as the first input data.

[0100] Step 3412: Obtain the increase value of the electricity consumption time series data and the accompanying factor time series data of the next moment compared to the previous moment within a preset long time period; obtain the electricity consumption time series growth rate data and the accompanying factor growth rate data of the preset long time period; and use the electricity consumption time series growth rate data and the accompanying factor growth rate data as the second input data.

[0101] Step 3413: Input the data sequence of each time series data of the first input data within a preset long time period into the sequence feature extraction unit, extract the time series features of each time series data, and characterize the time relationship features of each time series data;

[0102] Step 3414: Input the electricity consumption time-series growth rate data and the accompanying factor growth rate data of the second input data at the same time into the convolutional layer unit to obtain the spatial relationship features of the electricity consumption time-series growth rate data and the accompanying factor growth rate data.

[0103] Step 3415: Based on the temporal and spatial relationship features of the input data, the fully connected layer is used for analysis and judgment to determine whether the input data is electricity theft feature data.

[0104] In this embodiment, based on the concepts of channel attention and spatial attention mechanisms, the temporal and spatial relationship characteristics of electricity consumption time-series data and accompanying factor time-series data are analyzed. The relationship between electricity consumption time-series data and accompanying factor time-series data is mined from multiple dimensions. Based on the input multi-dimensional sequence data, a sequence feature extraction unit is used to extract the temporal features of each time-series data point within a preset long time period, representing the temporal relationship characteristics of each data point. Based on the multi-dimensional time-series growth rate sequence data, the spatial relationship between the electricity consumption time-series growth rate data and the accompanying factor growth rate data is mined using convolutional kernels. Based on the temporal and spatial relationship characteristics, binary classification analysis is performed. In this embodiment, the sequence feature extraction unit can be implemented based on an LSTM model, with multiple convolutional layers in the convolutional layer unit. Each convolutional layer includes multiple convolutional kernels of different scales to extract multi-scale spatial relationship characteristics.

[0105] In another embodiment, the neural network-based intelligent algorithm further includes: converting the electricity consumption time-series data curve and the accompanying factor time-series data curve based on a preset long time period into a two-dimensional image with time as the horizontal axis, wherein the two-dimensional image includes the electricity consumption time-series data curve and the accompanying factor time-series data curve with the same horizontal axis within the same coordinate system region.

[0106] Based on the input of multi-dimensional time-series data curve images into a convolutional neural network, the system identifies and mines the change characteristics and correlation characteristics of electricity consumption time-series data curves and accompanying factor time-series data curves. It identifies whether the distribution characteristics of electricity consumption time-series data and accompanying factor time-series data within a preset long time period in the image conform to the normal electricity consumption change characteristics. The convolutional neural network includes multiple convolutional layers, and each convolutional layer includes multiple non-one-dimensional convolutional kernels.

[0107] In this embodiment of the application, for the convolutional layer unit that acquires spatial relationship features, when the input is electricity consumption time-series growth rate data and accompanying factor growth rate data over a preset long period of time, in order to mine spatial relationship features, in this embodiment of the application, the electricity consumption time-series growth rate data and accompanying factor growth rate data at the same moment are input into the convolutional layer unit, and the convolution kernel uses one-dimensional convolution sliding to perform convolution operation on the input data.

[0108] For convolutional neural networks used in multi-dimensional time-series data curve images, in order to identify whether the distribution characteristics of electricity consumption time-series data and accompanying factor time-series data within a preset long time period in the image conform to normal electricity consumption change characteristics, multiple non-one-dimensional convolutional kernels are used in the convolutional layers, such as 3*3 convolutional kernels and 5*5 convolutional kernels. In this embodiment, considering that the time change trends of electricity consumption time-series data and accompanying factor time-series data can be classified into various types according to the electricity consumption habits of different users, for example, the electricity consumption of working people is relatively high in the morning, noon, and evening after get off work on weekdays, and relatively low at other times (showing an upward and downward trend during non-working hours, and decreasing at night when resting), while the electricity consumption of working people during the daytime on weekends is less variable. Therefore, using convolutional neural networks to identify the distribution characteristics of user electricity consumption time-series data and accompanying factor time-series data can identify whether it is a normal electricity consumption change characteristic; if not, it is electricity theft.

[0109] Furthermore, in this embodiment, the analysis result of inputting the time relationship features and spatial relationship features of the input data into the fully connected layer to analyze and judge whether the input data is electricity theft feature data is recorded as the first analysis result. The analysis result of inputting the multi-dimensional time series data curve image into the convolutional neural network to identify whether the distribution features of electricity consumption time series data and accompanying factor time series data within a preset long time period in the image conform to the normal electricity consumption change features is recorded as the second analysis result. The final electricity theft analysis result is determined by comprehensively deciding based on the first analysis result and the second analysis result.

[0110] This application provides an anti-electricity theft analysis system based on big data on electricity consumption and intelligent algorithm models, comprising:

[0111] The first electricity theft analysis unit is used to analyze the line loss data within the area to obtain the first electricity theft analysis result;

[0112] The electricity consumption data acquisition unit is used to acquire users' electricity consumption data and accompanying factor data. The electricity consumption data includes electricity consumption time-series data; the accompanying factor data includes weather, calendar, and household attribute data that accompany the user's electricity consumption at the same time.

[0113] The second electricity theft analysis unit is used to analyze the second electricity theft analysis results based on the user's electricity consumption data and accompanying factor data using an intelligent algorithm model.

[0114] The third electricity theft analysis unit is used to determine the existence of electricity theft based on the results of the first and second electricity theft analyses.

[0115] The electricity theft location unit is used to analyze and locate the location and time period of electricity theft based on line loss data, user electricity consumption data and accompanying factor data within the area when electricity theft is confirmed.

[0116] Specific limitations regarding the anti-electricity theft analysis system based on big data of electricity consumption and intelligent algorithm models can be found in the limitations of the anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm models mentioned above, and will not be repeated here. Each unit in the aforementioned anti-electricity theft analysis system based on big data of electricity consumption and intelligent algorithm models can be implemented entirely or partially through software, hardware, or a combination thereof. These units can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each unit.

[0117] This application provides an electronic device, the electronic device comprising:

[0118] processor;

[0119] Memory used to store processor-executable instructions;

[0120] The processor executes the executable instructions to implement the anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm models, as described above.

[0121] The processor may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the electronic device may optionally include input interfaces and output interfaces. The processor, memory, and input / output interfaces can be connected via a bus or signal lines. Various peripheral devices can be connected to the input / output interfaces via buses, signal lines, or circuit boards. The input / output interfaces can be used to connect at least one input / output related peripheral device to the processor and memory.

[0122] This application provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the steps of the aforementioned anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm models. For example, the computer-readable storage medium may be a read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage node, etc.

[0123] This invention is not limited to the specific embodiments described above. Any modifications made by those skilled in the art based on the above concept without creative effort are within the scope of protection of this invention.

Claims

1. A method for anti-electricity theft analysis based on big data on electricity consumption and intelligent algorithm models, characterized in that, Includes the following steps: The first electricity theft analysis result is based on the analysis of line loss data within the region; Acquire user electricity consumption data and accompanying factor data. The electricity consumption data includes time-series electricity consumption data. The accompanying factor data includes weather, calendar, and household attribute data that accompany the user's electricity consumption at the same time. Based on users' electricity consumption data and accompanying factor data, the results of the second electricity theft analysis are analyzed using an intelligent algorithm model. Based on the results of the first and second electricity theft analyses, it was determined that electricity theft had occurred. If electricity theft is confirmed, the location and time period of the theft are analyzed and determined based on line loss data, user electricity consumption data, and accompanying factor data within the area. The second electricity theft analysis, based on user electricity consumption data and accompanying factor data, utilizes an intelligent algorithm model to analyze the results, including: performing trend similarity analysis on the daily, weekly, and monthly electricity consumption time-series data of all users, grouping users with similar trends into one category; if the trends of daily, weekly, and monthly electricity consumption time-series data are all similar, the user is grouped into one category; if any one of the three trends is dissimilar, the user is not grouped into one category and is identified as a candidate electricity theft user; creating a separate category for users not grouped into any category; identifying users in categories with fewer than a first preset number as candidate electricity theft users; for candidate electricity theft users, acquiring their historical electricity consumption time-series data and historical accompanying factor time-series data; mining the user's electricity consumption patterns under the influence of accompanying factors based on the user's historical electricity consumption time-series data and historical accompanying factor time-series data; and analyzing whether the candidate electricity theft user has engaged in electricity theft behavior in the current time period based on the user's electricity consumption time-series data and accompanying factor time-series data, utilizing the user's electricity consumption patterns. The trend similarity analysis includes: calculating the growth rate of the daily electricity consumption time series data of the first user from the previous time to the next time, to obtain the daily electricity consumption time series growth rate value sequence of the first user; calculating the growth rate of the daily electricity consumption time series data of the second user from the previous time to the next time, to obtain the daily electricity consumption time series growth rate value sequence of the second user; calculating the difference between the daily electricity consumption time series growth rate value of the first user and the daily electricity consumption time series growth rate value of the second user at the same time based on the daily electricity consumption time series growth rate value sequence of the first user and the daily electricity consumption time series growth rate value of the second user, and obtaining the sum of the differences corresponding to all times in the daily electricity consumption time series data as the similarity between the trend of the daily electricity consumption time series data of the first user and the trend of the daily electricity consumption time series data of the second user; The method of mining user electricity consumption patterns under the influence of accompanying factors based on users' historical electricity consumption time-series data and historical accompanying factor time-series data includes: using historical electricity consumption time-series data and historical accompanying factor time-series data of different users as a first dataset, training a neural network model to mine the user's electricity consumption patterns under the influence of accompanying factors, denoted as the first electricity consumption pattern, and obtaining a first electricity consumption analysis model representing the first electricity consumption pattern; using the first electricity consumption analysis model representing the first electricity consumption pattern as the base model, using historical electricity consumption time-series data and historical accompanying factor time-series data of the same user as a second dataset, transforming the base model to mine the user's electricity consumption patterns under the influence of accompanying factors, denoted as the second electricity consumption pattern, and obtaining a second electricity consumption analysis model representing the second electricity consumption pattern.

2. The anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm model according to claim 1, characterized in that, The basic model is transformed to obtain a second electricity consumption analysis model that represents the second electricity consumption pattern, including: Based on the second sample input of the second dataset, the final encoding result of the basic model encoding the electricity consumption time series data and the accompanying factor time series data is obtained, which is the result before the output layer of the basic model. The final encoding result based on the basic model is transformed by the first linear transformation matrix to output the first encoding transformation result; Based on the first encoding transformation result, a nonlinear transformation is performed through a nonlinear activation function layer to obtain the nonlinear encoding result; Based on the nonlinear encoding result, a first linear inverse transformation matrix opposite to the first linear transformation matrix is ​​used to obtain the first decoding result. The final encoding result of the basic model and the first decoding result are combined as the fused first decoding result. The fused first decoding result is input into the output layer of the basic model for fully connected processing to obtain the electricity theft identification result of the second sample. The second electricity consumption analysis model is obtained by training the first linear transformation matrix, the nonlinear activation function layer, and the first linear inverse transformation matrix based on the error of the second sample electricity theft labeling data and the actual electricity theft identification results.

3. The anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm model according to claim 1, characterized in that, The step of creating a separate class based on users not assigned to any class, and identifying users in a class with fewer than a first preset value as candidate electricity theft users, also includes: Based on users in a class whose number of users is greater than a first preset value, determine the height distance between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curve of the other users in the class, and obtain the sum of the height distances between the electricity consumption time-series data curve of each user and the electricity consumption time-series data curves of all other users in the class; The sum of the height distances corresponding to the electricity consumption time-series data curves of each user are sorted from largest to smallest, and the electricity consumption time-series data curves of users whose values ​​are at the top of the sorted preset values ​​are selected as candidate electricity theft users.

4. The anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm model according to claim 1, characterized in that, The method of mining users' electricity consumption patterns under the influence of accompanying factors based on users' historical electricity consumption time-series data and historical accompanying factor time-series data includes: using users' historical electricity consumption time-series data and historical accompanying factor time-series data as input, training a neural network-based intelligent algorithm model to mine users' electricity consumption patterns, and characterizing users' electricity consumption patterns based on the trained neural network model. The neural network-based intelligent algorithm includes: The first input data is the electricity consumption time series data and the accompanying factor time series data over a preset long period. Obtain the increase value of the electricity consumption time series data and the accompanying factor time series data of the next moment compared to the previous moment within a preset long time period; obtain the electricity consumption time series growth rate data and the accompanying factor growth rate data of the preset long time period; and use the electricity consumption time series growth rate data and the accompanying factor growth rate data as the second input data. Each time series data of the first input data is input into the sequence feature extraction unit within a preset long time period to extract the time series features of each time series data, thereby characterizing the time relationship features of each time series data. The electricity consumption time-series growth rate data and the accompanying factor growth rate data at the same time as the second input data are input into the convolutional layer unit to obtain the spatial relationship features of the electricity consumption time-series growth rate data and the accompanying factor growth rate data. The temporal and spatial relationship features of the input data are fed into a fully connected layer for analysis and judgment to determine whether the input data is characteristic data of electricity theft.

5. A system for analyzing electricity theft based on big data on electricity consumption and an intelligent algorithm model, as described in any one of claims 1-4, characterized in that, include: The first electricity theft analysis unit is used to analyze the line loss data within the area to obtain the first electricity theft analysis result; The electricity data acquisition unit is used to acquire the user's electricity data and accompanying factor data, wherein the electricity data includes electricity consumption time-series data. The accompanying factor data includes weather, calendar, and household attribute data that accompany the user's electricity consumption at the same time; The second electricity theft analysis unit is used to analyze the second electricity theft analysis results based on the user's electricity consumption data and accompanying factor data using an intelligent algorithm model. The third electricity theft analysis unit is used to determine the existence of electricity theft based on the results of the first and second electricity theft analyses. The electricity theft location unit is used to analyze and locate the location and time period of electricity theft based on line loss data, user electricity consumption data and accompanying factor data within the area when electricity theft is confirmed.

6. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; The processor executes the executable instructions to implement the anti-electricity theft analysis method based on big data on electricity consumption and intelligent algorithm model as described in any one of claims 1-4.

7. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the instruction is executed by the processor, it implements the steps of the anti-electricity theft analysis method based on big data of electricity consumption and intelligent algorithm model as described in any one of claims 1-4.

Images