Electricity stealing characteristic analysis system for power consumer

Abstract

The invention discloses a power consumer electricity stealing characteristic analysis system, which comprises the following steps: classifying users, and collecting different data formats, update frequencies and data qualities; preprocessing the data, generating behavior characteristics including power consumption peak and valley periods and power consumption changes in holidays and festivals, and generating one-dimensional data and two-dimensional data; the method comprises the steps of training a model through historical data, establishing an electricity stealing characteristic analysis model, performing year-on-year and link-on-year analysis on electric quantity, line loss data and cover opening records, optimizing model performance through a cross validation method, mining internal association among data, further improving the detection rate of electricity stealing behaviors, and improving the detection efficiency. According to the system, through automatic data acquisition, intelligent data analysis and visual display, comprehensive monitoring of power consumption behaviors of power consumers and accurate identification of electricity stealing characteristics are realized, analysis results are visually displayed, users are helped to rapidly understand data, scientific decision basis is provided, and operation cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of power systems, and in particular to a system for analyzing characteristics of electricity theft by power users. Background Art

[0002] The power supply company is a bridge between power generation units and power users. That is, the electricity sent by power plants and other units needs to be transmitted to each power user through the power supply company. From the perspective of the power supply company, it realizes the company's profit by selling electricity. However, if the power user steals electricity, it will cause economic losses to the power supply company.

[0003] The theft of electricity by some or some users may affect the normal operation of the entire power grid, thus preventing other legitimate users from using electricity normally. In serious cases, the theft of electricity not only damages the economic interests of power companies, but may also cause power grid accidents, resulting in property losses for users or damage to the lives and health of related personnel.

[0004] The current electricity theft has caused huge losses to the power system. Traditional manual detection methods are inefficient and inaccurate, and cannot meet the needs of modern power systems for electricity theft detection. With the development of big data, cloud computing and artificial intelligence technologies, automated and intelligent electricity theft detection has become possible. These advanced technologies can be used to improve the efficiency and accuracy of electricity theft detection based on user electricity consumption data.

[0005] A Chinese patent discloses a method and system for monitoring electricity consumption (authorization announcement number CN 106645935 A). This patented technology can screen users in the monitored area, thereby avoiding monitoring all users in a certain area, improving monitoring efficiency, and can determine the way of electricity theft, providing a basis for manual judgment in the next step. However, the data source is single, lacks mutual comparison, and the result display is not intuitive enough. Therefore, the present invention provides a system for analyzing the characteristics of electricity theft by power users to solve the problems raised in the above background technology. Summary of the invention

[0006] The purpose of the present invention is to provide a power user electricity theft feature analysis system to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A power user electricity theft feature analysis system comprises the following steps:

[0009] Step 1: Classify users and collect data in different formats, update frequencies and data quality.

[0010] Step 2: Based on a deep understanding of electricity consumption behavior, pre-process the data to generate behavioral features, including changes in electricity consumption during peak and valley periods and holidays, and generate one-dimensional and two-dimensional data to better mine the potential information in the data.

[0011] Step 3: Use historical data to train the model, establish a feature analysis model for electricity theft, conduct year-on-year and month-on-month analysis on power consumption, line loss data, and cover opening records, use cross-validation methods to optimize model performance, and mine the inherent correlations between data, further improving the detection rate of electricity theft.

[0012] Preferably, users are classified according to their registration information, including residential users, large electricity users, large industrial users, ordinary industrial users, non-industrial users, commercial users and agricultural production users, to establish typical samples of various industries;

[0013] Preferably, the electricity theft feature analysis model output is performed according to the following method:

[0014] 1. Assume that the power consumption sequence of user X is X = {x1, x2, ..., x n}, use the following formula to normalize the sequence:

[0015]

[0016] x is the daily electricity consumption; x' is the relative value of daily electricity consumption; x max is the maximum daily power consumption; x min Generate a daily power consumption-time curve graph for the minimum daily power consumption.

[0017] Second, for the normalized power consumption sequence X', it is converted into a two-dimensional image through the MTF method to obtain an MTF matrix of dimension n×n.

[0018] 3. After converting the user's electricity consumption data into two dimensions, we perform convolution operations through the convolution kernel window, obtain a more abstract feature convolution matrix through in-depth analysis, and use the ResNet residual block and convolution kernel to extract the deep spatial features of the two-dimensional image. The spatial features in the image can better reflect the temporal characteristics of the original sequence, and the deep residual network is used to capture its changes, so as to analyze the user's electricity consumption behavior at multiple time scales.

[0019] Fourth, in order to enable the model to learn these two types of features at the same time, the output of the convolution module and the one-dimensional data are spliced ​​to generate a new set of fused features. The splicing operation is shown in the following formula:

[0020]

[0021]

[0022] Where: superscript 1D and 2D represent the one-dimensional and two-dimensional feature tensors of variable information respectively; T joint is the fusion feature after splicing; T i 1D 、T i 2D are the i-th element of the one-dimensional and two-dimensional feature tensors respectively; f joint () is the concatenation function, which concatenates two tensors along the same connection axis; m and n are the corresponding dimensions of the two at the connection axis.

[0023] In order to enable the hybrid module to learn the feature information from the dual channels fairly, n and m are both set to 128. The concatenation operation integrates the two types of power consumption feature information extracted by the convolution module, and then inputs this new set into the fully connected layer with 64 neurons for nonlinear feature learning. The fully connected layer is used to classify unfamiliar daily power consumption to detect the changes in daily power consumption caused by various types of power theft. At this time, those with a power theft probability greater than 0.5 are judged as power theft users, otherwise they are judged as normal users.

[0024] Preferably, the MTF is a method for converting time series data into spatial image data. In the MTF field, each pixel represents a state, and the conversion relationship between states is modeled by a Markov chain.

[0025] Preferably, the collection frequency can be set. A reasonable collection frequency can ensure timely acquisition of the latest data while avoiding excessive pressure on the data source. The data includes key information such as household number, household name, user classification, and whether there is a history of electricity theft.

[0026] Preferably, the preprocessing of the data includes cleaning, conversion and integration to ensure the accuracy and consistency of the data.

[0027] Preferably, the historical data contains rich electricity usage information and potential electricity theft patterns. By learning from these data, the electricity theft feature analysis model can gradually grasp the characteristics of electricity theft behavior.

[0028] Preferably, the cover opening record includes information such as the time, frequency, and location of the cover opening, and combined with advanced data analysis algorithms and machine learning technology, the potential correlation between the cover opening record and the power consumption and line loss data is mined. By analyzing the geographical location information of the cover opening record, it can also be found whether there is an abnormal concentration of cover opening records in a certain area. If so, there may be electricity theft gangs or management loopholes in the area, and supervision and inspection efforts need to be further strengthened.

[0029] Preferably, cross-validation uses time series cross-validation. Time series data is data collected at different time points. The data is split into training sets and validation sets according to time, and different validation modules are generated quarterly. The uniformity of power consumption, line loss data and cover opening records is combined, and then the accuracy is checked. This can be used to optimize model performance, evaluate the accuracy and generalization ability of the model, and ensure that the model can accurately identify electricity theft in practical applications.

[0030] Preferably, year-on-year / month-on-month statistical analysis of historical electricity consumption of users: calculate the year-on-year and month-on-month changes in historical electricity consumption of users, and intuitively present the quantitative differences in electricity consumption with a bar chart, and clearly show the trend of electricity consumption changes with a discount chart. By analyzing these graphics, it is possible to keenly capture abnormal fluctuations in user electricity consumption, such as a sudden sharp increase or decrease in the electricity consumption of users who were originally stable, which may indicate the existence of electricity theft.

[0031] Optimize the year-on-year and month-on-month analysis of the user's historical power loss data: Accurately obtain the year-on-year and month-on-month information of the user's historical power loss data, and display the changes in line loss data in the form of a line graph, so that the fluctuation trend of line loss is clear at a glance. Excessive line loss rate may be caused by power theft, so close attention to line loss data can help discover potential power theft problems.

[0032] Preferably, the results are visualized, including heat maps of electricity consumption in specific areas, to make the data more intuitive and easy to understand. Visualization can help users quickly find anomalies and trends in the data, facilitating in-depth analysis and decision-making. Customized visualization can meet the needs of different users and improve user experience and satisfaction with the system.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1. The system realizes comprehensive monitoring of power users' electricity consumption behavior and accurate identification of electricity theft characteristics through automated data collection, intelligent data analysis and visual display.

[0035] 2. Automated data collection: reduce manual intervention and improve the efficiency and accuracy of data collection.

[0036] 3. Data drilling and analysis: Use advanced algorithms to deeply mine data and improve the accuracy of identifying electricity theft characteristics.

[0037] 4. Visual display: intuitively display the analysis results to help users quickly understand the data.

[0038] 5. Decision support: provide scientific decision-making basis and reduce operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1The figure is a schematic diagram of a system for analyzing the characteristics of electricity theft by power users. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] See also Figure 1 In an embodiment of the present invention, a power user electricity theft feature analysis system comprises the following steps:

[0042] Step 1: Classify users and collect data in different formats, update frequencies and data quality.

[0043] Step 2: Based on a deep understanding of electricity consumption behavior, pre-process the data to generate behavioral features, including changes in electricity consumption during peak and valley periods and holidays, and generate one-dimensional and two-dimensional data to better mine the potential information in the data.

[0044] Step 3: Use historical data to train the model, establish a feature analysis model for electricity theft, conduct year-on-year and month-on-month analysis on power consumption, line loss data, and cover opening records, use cross-validation methods to optimize model performance, and mine the inherent correlations between data, further improving the detection rate of electricity theft.

[0045] Users are classified according to their registration information into residential users, large electricity users, large industrial users, ordinary industrial users, non-industrial users, commercial users and agricultural production users, and typical samples of various industries are established;

[0046] The electricity theft feature analysis model output is as follows:

[0047] 1. Assume that the power consumption sequence of user X is X = {x1, x2, ..., x n}, use the following formula to normalize the sequence:

[0048]

[0049] x is the daily electricity consumption; x' is the relative value of daily electricity consumption; x max is the maximum daily power consumption; x min Generate a daily power consumption-time curve graph for the minimum daily power consumption.

[0050] Second, for the normalized power consumption sequence X', it is converted into a two-dimensional image through the MTF method to obtain an MTF matrix of dimension n×n.

[0051] 3. After converting the user's electricity consumption data into two dimensions, we perform convolution operations through the convolution kernel window, obtain a more abstract feature convolution matrix through in-depth analysis, and use the ResNet residual block and convolution kernel to extract the deep spatial features of the two-dimensional image. The spatial features in the image can better reflect the temporal characteristics of the original sequence, and the deep residual network is used to capture its changes, so as to analyze the user's electricity consumption behavior at multiple time scales.

[0052] Fourth, in order to enable the model to learn these two types of features at the same time, the output of the convolution module and the one-dimensional data are spliced ​​to generate a new set of fused features. The splicing operation is shown in the following formula:

[0053]

[0054]

[0055] Where: superscript 1D and 2D represent the one-dimensional and two-dimensional feature tensors of variable information respectively; T joint is the fusion feature after splicing; T i 1D 、T i 2D are the i-th element of the one-dimensional and two-dimensional feature tensors respectively; f joint () is the concatenation function, which concatenates two tensors along the same connection axis; m and n are the corresponding dimensions of the two at the connection axis.

[0056] In order to enable the hybrid module to learn the feature information from the dual channels fairly, n and m are both set to 128. The concatenation operation integrates the two types of electricity feature information extracted by the convolution module, and then inputs this new set into the fully connected layer with 64 neurons for nonlinear feature learning. At this time, those with a probability of electricity theft greater than 0.5 are judged as electricity theft users, otherwise they are judged as normal users.

[0057] The MTF is a method for converting time series data into spatial image data. In the MTF field, each pixel represents a state, and the transition relationship between states is modeled by a Markov chain.

[0058] The collection frequency can be set. A reasonable collection frequency can ensure timely acquisition of the latest data while avoiding excessive pressure on the data source. The data includes key information such as household number, household name, user classification, and whether there is a history of electricity theft.

[0059] Data preprocessing includes cleaning, transformation and integration to ensure data accuracy and consistency.

[0060] Monitoring the characteristics of electricity usage behavior reflects the user's electricity usage habits and patterns and discovers abnormal electricity usage behavior. By analyzing these characteristics, the accuracy of electricity theft feature extraction can be further improved, providing more valuable input for subsequent model training and decision-making.

[0061] The historical data contains rich electricity usage information and potential electricity theft patterns. By learning from these data, the electricity theft feature analysis model can gradually grasp the characteristics of electricity theft behavior.

[0062] The opening records include information such as the time, frequency, and location of the opening. Combined with advanced data analysis algorithms and machine learning technology, the potential correlation between the opening records and power consumption and line loss data is explored. By analyzing the geographical location information of the opening records, it is also possible to find out whether there is an abnormal concentration of opening records in a certain area. If so, there may be electricity theft gangs or management loopholes in the area, and further supervision and inspection efforts need to be strengthened.

[0063] Cross-validation uses time series cross-validation. Time series data is data collected at different time points. The data is split into training sets and validation sets according to time, and different validation modules are generated quarterly. The uniformity of power consumption, line loss data and cover opening records is combined, and then the accuracy is checked. It can be used to optimize model performance, evaluate the accuracy and generalization ability of the model, and ensure that the model can accurately identify electricity theft in practical applications.

[0064] The working principle of the present invention is:

[0065] Year-on-year / month-on-month statistical analysis of users' historical electricity consumption: Calculate the year-on-year and month-on-month changes in users' historical electricity consumption, and use a bar chart to visually present the quantitative differences in electricity consumption, and use a discount chart to clearly show the trend of electricity consumption changes. By analyzing these graphics, it is possible to keenly capture abnormal fluctuations in users' electricity consumption, such as a sudden sharp increase or decrease in users with stable electricity consumption, which may indicate electricity theft.

[0066] Year-on-year and month-on-month analysis of users' historical power loss data: Accurately obtain year-on-year and month-on-month information on users' historical power loss data, and display the changes in line loss data in the form of a line graph, so that the fluctuation trend of line loss is clear at a glance. Excessive line loss rate may be caused by power theft, so close attention to line loss data can help discover potential power theft problems.

[0067] By analyzing and summarizing a large amount of user electricity consumption data, we can help extract the graphical features of characteristic electricity theft data, thereby providing a strong basis for the analysis and judgment of user electricity consumption data.

[0068] Visualizing the results, including heat maps of power consumption in specific areas, can make the data more intuitive and easy to understand. Visualization can help users quickly find anomalies and trends in the data, facilitating in-depth analysis and decision-making. Customized visualization can meet the needs of different users and improve user experience and satisfaction with the system.

[0069] In summary, a power user electricity theft feature analysis system is used to improve the efficiency and accuracy of anti-electricity theft work and ensure the safe and stable operation of the power system.

[0070] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A power user electricity theft feature analysis system, characterized in that: Including the electricity theft feature analysis model, the system analysis process includes the following steps: Step 1: Classify users and collect different data formats, update frequencies and data quality; Step 2: Preprocess the data to generate behavioral features, including peak and valley periods and changes in electricity consumption during holidays, and generate one-dimensional and two-dimensional data; Step 3: Use historical data to train the model, establish a feature analysis model for electricity theft, conduct year-on-year and month-on-month analysis on power consumption, line loss data, and cover opening records, use cross-validation methods to optimize model performance, and mine the inherent correlations between data, further improving the detection rate of electricity theft.

2. A power user electricity theft feature analysis system according to claim 1, characterized in that: Users are classified according to their registration information, consisting of residential users, large electricity users, large industrial users, ordinary industrial users, non-industrial users, commercial users and agricultural production users, to establish typical samples for each industry.

3. The power user electricity theft characteristic analysis system according to claim 1 is characterized in that: The electricity theft feature analysis model output is as follows:

1. Assume that the power consumption sequence of user X is X = {x1, x2, ..., x n }, use the following formula to normalize the sequence: x is the daily electricity consumption; x' is the relative value of daily electricity consumption; x max is the maximum daily power consumption; x min Generate a curve chart of daily power consumption-time for the minimum daily power consumption; Second, for the normalized power consumption sequence X', convert it into a two-dimensional image through the MTF method to obtain an MTF matrix of dimension n×n; 3. After converting the user's electricity consumption data into two dimensions, the convolution operation is performed through the convolution kernel window. Through in-depth analysis, a more abstract feature convolution matrix is ​​obtained, and the deep spatial features of the two-dimensional image are extracted using the ResNet residual block and the convolution kernel. The spatial features in the image can better reflect the temporal features of the original sequence. The deep residual network can capture its changes and analyze the user's electricity consumption behavior at multiple time scales. Fourth, in order to enable the model to learn these two types of features at the same time, the output of the convolution module and the one-dimensional data are spliced ​​to generate a new set of fused features. The splicing operation is shown in the following formula: Where: superscript 1D and 2D represent the one-dimensional and two-dimensional feature tensors of variable information respectively; T joint is the fusion feature after splicing, T i 1D , T i 2D are the i-th element of the one-dimensional and two-dimensional feature tensors respectively; f joint () is the concatenation function, which concatenates two tensors along the same connection axis; m and n are the corresponding dimensions of the two at the connection axis; This new set is input into a fully connected layer with 64 neurons for nonlinear feature learning. The fully connected layer is used to classify unfamiliar daily electricity consumption to detect changes in daily electricity consumption caused by various types of electricity theft. At this time, those with a probability of electricity theft greater than 0.5 are judged as electricity theft users, otherwise they are judged as normal users.

4. The power user electricity theft characteristic analysis system according to claim 1 is characterized in that: The collection frequency can be set, and the data includes key information such as household number, household name, user classification, and whether there is a history of electricity theft.

5. The power user electricity theft feature analysis system according to claim 1, characterized in that: Data preprocessing includes cleaning, transformation and integration to ensure data accuracy and consistency.

6. The power user electricity theft characteristic analysis system according to claim 1 is characterized in that: The opening record includes information such as the time, frequency, and location of the opening.

7. The power user electricity theft characteristic analysis system according to claim 1 is characterized in that: Cross-validation uses time series cross-validation. Time series data is data collected at different time points. The data is split into training sets and validation sets according to time. Different validation modules are generated quarterly, combining the uniformity of power consumption, line loss data and cover opening records, and then checking the accuracy.

8. The power user electricity theft feature analysis system according to claim 1, characterized in that: Year-on-year and month-on-month analysis of users' historical electricity consumption: Calculate the year-on-year and month-on-month changes in users' historical electricity consumption, and visually present the quantitative differences in electricity consumption with a bar chart, and clearly show the changing trend of electricity consumption with a discount chart.

9. The power user electricity theft feature analysis system according to claim 1, characterized in that: Year-on-year and month-on-month analysis of users' historical line loss data: obtain year-on-year and month-on-month information of users' historical line loss data, and display the changes in line loss data in the form of a line graph.

10. The power user electricity theft feature analysis system according to claim 1, characterized in that: The system provides a visual display of the results, including heat maps of electricity consumption in specific areas.

Images