Method and Platform for Generating Massive Label Data, Electricity Theft Identification Method and System

Abstract

The present invention discloses a method and platform for generating a large amount of tag data, a method and system for electricity theft identification, including obtaining first electricity consumption data of a user, and training a pre-established control model by using the first electricity consumption data to extract electricity consumption behavior characteristics corresponding to the first electricity consumption data; building an electricity consumption simulation platform, inputting the first electricity consumption data and its corresponding electricity consumption behavior characteristics into the electricity consumption simulation platform, controlling the load in the electricity consumption simulation platform, amplifying and outputting simulated electricity consumption data; applying at least one electricity theft means to interfere with the metering of the simulated electricity consumption data at the output end of the electricity consumption simulation platform, and outputting second electricity consumption data interfered by the electricity theft means; extracting data characteristics of the second electricity consumption data, associating the data characteristics with the corresponding electricity theft means, and obtaining a number of tag data to provide data support for realizing electricity theft discrimination and load prediction research.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal power consumption discrimination, and particularly to a method and platform for generating a large amount of tag data, a power theft identification method and system. Background Art

[0002] Existing power theft identification usually involves establishing an identification model, training the identification model based on user power consumption data, and then accurately identifying power theft means. However, due to privacy and confidentiality issues regarding user power consumption data, the power grid company can provide limited user power consumption data, which cannot meet the sample size standard required for training the model, resulting in an unbalanced data set during the model training process. In addition, the data types and collection methods collected by power grids in different provinces vary, leading to differences in data quality and data missing problems.

[0003] Currently, power consumption data is usually collected through two methods: hardware and software. For hardware, both electromechanical integrated meters and fully electronic meters in smart meters can achieve power consumption data collection. However, electromechanical integrated meters cannot collect and analyze data with lost pulses during power consumption data collection. Although fully electronic meters already have analysis technologies for preventing abnormal power consumption, the equipment repair of fully electronic meters is complex and expensive, lacking universality and unable to provide a large amount of available data for data analysis. At the same time, for software, although a large amount of data generated by user power consumption behavior can be obtained through non-intrusive load monitoring, the traditional non-intrusive user load monitoring system has low accuracy in monitoring peak and off-peak loads, the monitoring level is not ideal, and there are noise data in the monitored abnormal power consumption data, with uneven data quality and unable to be directly used for the analysis of power theft behavior. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and platform for generating a large amount of tag data, a power theft identification method and system, so as to solve the problems in the prior art that power consumption data with lost pulses cannot be collected and the quality of the collected power consumption data is not high.

[0005] To achieve the above object, the first aspect of the present invention provides a method for generating a large amount of tag data, including the following steps:

[0006] S1: Obtain the first power consumption data of a user, and use the first power consumption data to train a pre-established control model to extract power consumption behavior characteristics corresponding to the first power consumption data;

[0007] S2: Build a power consumption simulation platform, input the first power consumption data and its corresponding power consumption behavior characteristics into the power consumption simulation platform, control the load in the power consumption simulation platform, and amplify and output simulated power consumption data;

[0008] S3: Apply at least one electricity theft means to the output end of the electricity consumption simulation platform to interfere with the measurement of the simulated electricity consumption data, and output the second electricity consumption data after being interfered by the electricity theft means.

[0009] S4: Extract the data features of the second electricity consumption data, associate the data features with the corresponding electricity theft means, and obtain a number of labeled data.

[0010] Further, in step S1, step S1 includes the following sub-steps:

[0011] S101: Obtain the first electricity consumption data of the user, where the first electricity consumption data includes the existing historical electricity consumption data and the random electricity consumption data generated based on the historical electricity consumption data.

[0012] S102: Divide the first electricity consumption data into a training data set and a test data set, and input the training data set into a pre-established control model to perform iterative training on the control model, optimize the model parameters, and obtain a trained control model.

[0013] S103: Input the test data set as the first electricity consumption data into the trained control model, and extract the electricity consumption behavior features corresponding to the first electricity consumption data.

[0014] Further, in step S101, the random electricity consumption data is generated based on the random walk algorithm, and the specific method is:

[0015] Construct a user electricity consumption behavior sequence network according to the existing historical electricity consumption data, and establish a random walk model based on python. The random walk model takes each network node in the user electricity consumption behavior sequence network as a starting point, performs random walks respectively, generates a random walk sequence corresponding to each walk, and outputs each random data in each random walk sequence. The random data is the random electricity consumption data.

[0016] Further, the electricity consumption behavior features at least include daily electricity consumption, daily maximum / minimum load, average daily load, peak-valley electricity consumption, peak-valley average load, peak-valley difference, night load rate, electricity peak duration, and load curve.

[0017] Further, in step S2, the electricity consumption simulation platform includes an analog circuit and an amplifier circuit electrically connected to the analog circuit. The analog circuit includes multiple groups of loads with adjustable parameters, and the current and voltage of the loads are reduced by a preset multiple in proportion compared with the electrical equipment. The loads are used to simulate the electrical equipment; the amplifier circuit is used to amplify the current and voltage at the output end of the analog circuit by the preset multiple in proportion.

[0018] Further, a plurality of smart meters electrically connected to the amplifier circuit are provided at the output end of the power consumption simulation platform, and the specific method of step S3 is as follows:

[0019] Apply a power theft means to each smart meter according to the category of the power theft means, receive simulated power consumption data through the smart meter, cause the power theft means to interfere with the measurement of the simulated power consumption data, and output second power consumption data generated under different power theft means through each smart meter.

[0020] Further, step S4 includes the following sub-steps:

[0021] S401: Obtain the second power consumption data and its corresponding power consumption behavior characteristics to form a feature matrix;

[0022] S402: Use the principal component analysis method to calculate the principal components of the feature matrix and the cumulative contribution rate of each principal component in the power consumption behavior characteristics;

[0023] S403: Extract the principal components with a cumulative contribution rate greater than a preset condition according to the cumulative contribution rate of the principal components to form the data characteristics of the second power consumption data;

[0024] S404: Associate the data characteristics with the corresponding power theft means to obtain a number of tag data carrying power theft means information.

[0025] The second aspect of the present invention provides a massive tag data generation platform, including:

[0026] A control module, configured to extract power consumption behavior characteristics according to the first power consumption data, and generate a control signal based on the first power consumption data and the power consumption behavior characteristics, where the first power consumption data includes existing historical power consumption data and random power consumption data generated based on the historical power consumption data;

[0027] A power consumption simulation platform, configured to simulate the operating conditions of the electrical equipment according to the control signal, output and amplify the simulated power consumption data generated by the power consumption simulation platform;

[0028] A data acquisition and interference module, configured to collect the simulated power consumption data and apply a power theft means to interfere with the measurement of the simulated power consumption data, and output the second power consumption data after being interfered by the power theft means;

[0029] A data processing module, configured to extract the data characteristics of the second power consumption data and generate a number of tag data carrying power theft means information.

[0030] The third aspect of the present invention provides a power theft identification method, including the following steps:

[0031] P1: Generate a number of tag data carrying electricity theft means information by using the above-mentioned massive tag data generation method;

[0032] P2: Input the tag data and the corresponding electricity theft means into a pre-established electricity theft recognition model and perform iterative training on the electricity theft recognition model to obtain a trained electricity theft recognition model;

[0033] P3: Use the trained electricity theft recognition model to identify electricity theft means.

[0034] The fourth aspect of the present invention provides an electricity theft recognition system, including:

[0035] A massive tag data generation platform for generating a number of tag data carrying electricity theft means information based on a massive tag data generation platform;

[0036] A model training module for training an electricity theft recognition model according to the tag data and the corresponding electricity theft means information; and

[0037] An electricity theft recognition model for identifying electricity theft means according to the electricity consumption data output by the smart meter configured at the electricity user.

[0038] Based on the random walk algorithm, the present invention can generate a large number of random electricity consumption data on the basis of the existing historical electricity consumption data. The historical electricity consumption data and the random electricity consumption data can be used to train a control model, so that the trained control model can extract the electricity consumption behavior characteristics in the historical electricity consumption data and the random electricity consumption data, and then control an electricity consumption simulation platform. At the same time, by changing the load parameters and working hours in the electricity consumption platform, the electricity consumption conditions of the electrical equipment in various electricity consumption environments can be simulated. Finally, the second electricity consumption data of the electricity consumption simulation platform is collected by the smart meter with electricity theft means applied, and then the data characteristics of the second electricity consumption data are extracted. After associating the data characteristics with the electricity theft means, a large amount of tag data can be obtained, which makes up for the lack of massive data of abnormal electricity consumption. And each smart meter corresponds to an electricity theft means, which can effectively solve the problem of poor data quality and a large amount of noise data, and thus provide data support for realizing electricity theft discrimination and load prediction research. Description of the Drawings

[0039] Figure 1 It is a structural block diagram of the massive tag data generation platform according to Embodiment 1 of the present invention.

[0040] Figure 2 is Figure 1 The equivalent circuit diagram of part of the load in.

[0041] Figure 3 is Figure 1 The circuit structure diagram of the amplifier circuit in.

[0042] Figure 4 This is a flowchart of the method for generating a large amount of tag data in Embodiment 2 of the present invention.

[0043] Figure 5 It is Figure 4 the flowchart of step S1 in

[0044] Figure 6 It is Figure 4 the flowchart of step S4 in

[0045] Figure 7 This is a flowchart of the electricity theft identification method in Embodiment 3 of the present invention.

[0046] Figure 8 This is a structural block diagram of the electricity theft identification system in Embodiment 4 of the present invention. Specific embodiments

[0047] The following is a further detailed description through specific embodiments:

[0048] Embodiment 1

[0049] As Figure 1 shown, this is a structural block diagram of the large amount of tag data generation platform in this embodiment. The large amount of tag data generation platform in this embodiment includes a control module 1, a power consumption simulation platform 2, a data acquisition and interference module 3, and a data processing module 4 that are connected in sequence. The control module 1 extracts the power consumption behavior characteristics of the user according to the first power consumption data, and generates a control signal based on the first power consumption data and the power consumption behavior characteristics. Among them, the first power consumption data includes the existing historical power consumption data and the random power consumption data generated based on the historical power consumption data; the power consumption simulation platform 2 simulates the operating conditions of the power consumption equipment under various power consumption behaviors under the control of the control signal, and outputs and amplifies the generated simulated power consumption data; the data acquisition and interference module 3 acquires the simulated power consumption data generated by the power consumption simulation platform 2 and applies at least one electricity theft means to interfere with the measurement of the simulated power consumption data, and respectively outputs a plurality of second power consumption data after being interfered by each electricity theft means; the data processing module 4 extracts the data characteristics of the second power consumption data and generates several tag data carrying electricity theft means information.

[0050] The control module 1 is internally provided with a data acquisition module 11, a random walk model 12, and a control model 13. Among them:

[0051] The data acquisition module 11 is used to acquire the existing historical power consumption data of the user. In this embodiment, the historical power consumption data includes the historical power consumption data of residents, schools, and enterprises provided by the State Grid.

[0052] The random walk model 12 is used to generate a large amount of random power consumption data based on the historical power consumption data, and integrate it with the historical power consumption data to form the first power consumption data. In this embodiment, the random walk model 12 is established based on Python. When generating random data, it first constructs the existing historical power consumption data into a user power consumption behavior sequence network, and each historical power consumption data is a network node in the power consumption behavior sequence network; then, it performs random walks starting from each network node respectively to generate random walk sequences corresponding to each walk; finally, it outputs each random data in each random walk sequence, and the random data is the random power consumption data, so as to amplify the first power consumption data, expand the training samples when training the control model 13, and further improve the training accuracy of the control model 13.

[0053] The control model 13 is used to extract and output the power consumption behavior characteristics corresponding to the first power consumption data as the control signal to be input into the subsequent power consumption simulation platform 2, so as to realize the simulation control of the power consumption simulation platform 2. In this embodiment, the control model 13 adopts a convolutional neural network, and establishes an appropriate number of neuron computing nodes and a multi-layer operation hierarchical structure. The control model 13 takes the first power consumption data as the input for model training and optimizes the model parameters, and its output is the power consumption behavior characteristics corresponding to the first power consumption data. Specifically, the first power consumption data is divided into a training data set and a test data set according to a certain ratio, and the training data set is input into the convolutional neural network. The neuron computing nodes of each layer of the convolutional neural network calculate the first power consumption data, and output the power consumption behavior characteristics corresponding to the first power consumption data. The optimal input layer and output layer are selected, and the corresponding power consumption behavior characteristics are backpropagated to the convolutional neural network to iteratively optimize the control model 13 until the defined loss function tends to be stable or reaches the maximum number of iterations, so as to obtain the trained control model 13, and further obtain the mapping relationship between the input (i.e., the first power consumption data) and the output (i.e., the power consumption behavior characteristics), so as to realize the extraction of the user power consumption behavior characteristics.

[0054] The power consumption simulation platform 2 includes a simulation circuit 21 and an amplifier circuit 22 electrically connected to the simulation circuit 21. Among them:

[0055] The simulation circuit 21 can simulate the operating conditions of electrical appliances under various power consumption behaviors under the action of the control signal output by the control module 1. In this embodiment, the electrical appliances at least include refrigerators, TVs, air conditioners, floor heating, water heaters, fans, washing machines, computers, hair dryers, etc. Since different electrical appliances have different equivalent impedances, when simulating each electrical appliance, a combination of resistors, inductors, and / or capacitors can be used to form a load with different impedances and adjustable parameters (the equivalent circuits of some loads are as follows Figure 2As shown, different loads can be used to simulate different electrical equipment. The impedance of the load is reduced by a preset multiple (such as 10 times) in proportion to the equivalent impedance of the corresponding electrical equipment, so that the current and voltage output by the simulation circuit 21 are reduced by the preset multiple in proportion to the actual operating conditions of the electrical equipment, thereby realizing the simulation of the operating conditions of the electrical equipment. In this embodiment, the actual operating conditions of the electrical equipment are simulated by a load whose impedance is reduced in proportion to the equivalent impedance of the electrical equipment, which can reduce the power consumption of the entire electrical simulation platform 2 and achieve the purpose of energy conservation.

[0056] In this embodiment, the purpose of setting the parameters of the load to be adjustable is to increase the manual intervention in the operation of the load in the electrical simulation platform 2 by adjusting the parameters of the load, so that the influence of climate parameters, harmonic characteristics, and user switching behavior on the load can be increased during the process of the simulation circuit 21 simulating the operating conditions, making the simulated operating conditions of the simulation circuit 21 closer to the actual operating conditions of the electrical equipment. Specifically, for climate parameters, since different seasons and weather conditions will affect the usage frequency, duration, and usage time period of different electrical equipment. For example, summer is the high-frequency usage season for air conditioners, and the power consumption of TVs, computers, and hair dryers will increase significantly during typhoons and rainy days. While winter is the high-frequency usage season for electrical equipment such as floor heating. By specifically adjusting or switching the parameters of the load representing each electrical equipment, the climate parameter intervention of the simulation circuit 21 can be achieved; for harmonic characteristics, since harmonics will increase the power loss of the simulation circuit 21 and make the output simulated electrical data inaccurate, the simulation circuit 21 can be intervened by adjusting the load parameters or adding other loads; for user switching behavior, since the user's electricity consumption habits have a significant impact on the usage time period, usage duration, and usage frequency of electrical equipment, the user switching behavior can also be achieved by specifically adjusting or switching the parameters of the load representing each electrical equipment to intervene in the simulation circuit 21.

[0057] The amplifier circuit 22 is used to amplify the current and voltage output by the simulation circuit 21 by a preset multiple of equal ratio reduction, so that the current and voltage (simulated electrical data) finally output by the electrical simulation platform 2 are the same as the actual operating conditions of the electrical equipment, thereby enabling effective subsequent data collection.

[0058] Such as Figure 3As shown, the amplifier circuit 22 includes a triode Q1, a capacitor C3, a capacitor C4, a capacitor C5, a resistor R4, a resistor R5, a resistor R6, and a resistor R7. The capacitor C3 and the resistor R4 are connected in series between the positive and negative poles of the input end of the amplifier circuit 22. The capacitor C3 is adjacent to the positive pole of the input end while the resistor R4 is adjacent to the negative pole of the input end, and one end of the resistor R4 near the negative pole of the input end is grounded; the gate of the triode Q1 is electrically connected between the capacitor C3 and the resistor R4, the source of the triode Q1 is grounded through the resistor R5, and the drain of the triode Q1 forms the output end of the amplifier circuit 22 through the resistor R6; one end of the capacitor C4 is electrically connected between the drain of the triode Q1 and the resistor R6, and the other end of the capacitor C4 is grounded through the resistor R7; one end of the capacitor C5 is connected between the source of the triode Q1 and the resistor R5, and the other end of the capacitor C5 is grounded to achieve an equal ratio amplification of the current and voltage output by the analog circuit 21.

[0059] The data acquisition and interference module 3 includes at least one smart meter 31 electrically connected to the amplifier circuit 22 for acquiring the analog power consumption data output by the amplifier circuit 22; the smart meter 31 is modified, and a power theft means is applied to each smart meter 31 respectively to interfere with the counting of the smart meter 31, so that the second power consumption data after interference output by it is not equal to the analog power consumption data. In this embodiment, the power theft means at least includes stealing electricity by tampering with the metering program of the smart meter 31, stealing electricity by soldering a resistor on the circuit board of the smart meter 31, stealing electricity by changing the current sampling circuit of the smart meter 31, stealing electricity by applying carbon powder to the metering chip of the smart meter 31, stealing electricity by changing the voltage sampling circuit of the smart meter 31, and stealing electricity by damaging the carrier module rear partition of the smart meter 31, etc. Each smart meter 31 corresponds to a power theft means. Thus, the second power consumption data corresponding to each power theft means can be output.

[0060] The data processing module 4 receives the second power consumption data output by each smart meter 31 and extracts the data characteristics of the second power consumption data. Specifically, in this embodiment, the data processing module 4 uses the principal component analysis method to determine and extract the data characteristics with the cumulative contribution rate within the preset condition range among the data characteristics of the second power consumption data, associates them with the corresponding power theft means, and obtains a large amount of labeled data carrying power theft means information to provide data support for further realizing power theft discrimination and load prediction research.

[0061] The massive label data generation platform of this embodiment can generate a large number of random power consumption data with the same distribution as the historical power consumption data based on the limited historical power consumption data by setting a random walk model 12, so as to realize the amplification of the training samples of the control model 13, improve the training accuracy of the control model 13, enable it to accurately extract the power consumption behavior characteristics, and use them as the control signals of the power consumption simulation platform 2 to control the operating conditions of the load simulation power consumption equipment. At the same time, an intelligent electric meter 31 intervened by electricity stealing means is set at the end to collect the simulated power consumption data of the power consumption simulation platform 2, output the second power consumption data under the interference of the electricity stealing means, and after extracting the data characteristics, a large amount of label data carrying electricity stealing means information can be obtained to make up for the problems of the lack of massive abnormal power consumption data, few data samples and low quality.

[0062] Embodiment 2

[0063] As Figure 4 shown, it is a flowchart of the massive label data generation method of this embodiment. The massive label data generation method of this embodiment is implemented based on the massive data label generation platform of Embodiment 1, and includes a control module 1, a power consumption simulation platform 2, a data collection and interference module 3, and a data processing module 4 that are the same or similar in structure and function as those in Embodiment 1 to realize the collection of massive label data. Specifically, this embodiment includes the following steps:

[0064] S1: Obtain the first power consumption data and extract the power consumption behavior characteristics.

[0065] First, obtain the historical power consumption data of users announced by the State Grid, and generate a large number of random power consumption data based on the historical power consumption data, and integrate the historical power consumption data and the random power consumption data to form the first power consumption data; then, use the first power consumption data to train a pre-established control model 13, and the trained control model 13 can extract the power consumption behavior characteristics corresponding to the first power consumption data; finally, integrate the first power consumption data and the corresponding power consumption behavior characteristics to form a digital control signal. In this embodiment, the power consumption behavior characteristics are used to characterize the power consumption habits of users, which are manifested by the data size, data generation time, etc. of the first power consumption data. Specifically, the power consumption behavior characteristics at least include daily power consumption, daily maximum / minimum load, average daily load, peak-valley power consumption, peak-valley average load, peak-valley difference, night load rate, power consumption peak duration, and load curve, etc.

[0066] As Figure 5 shown, the step S1 includes the following sub-steps:

[0067] S101: Obtain the first power consumption data.

[0068] The historical electricity consumption data of residents, schools and enterprises publicly disclosed by the country is obtained through the data acquisition module 11, and a large number of random electricity consumption data with the same distribution as the historical electricity consumption data are randomly generated on the basis of the historical electricity consumption data by using the random walk model 12, forming the first electricity consumption data including the historical electricity consumption data and the random electricity consumption data.

[0069] In this embodiment, the random walk model 12 is established based on python; specifically, the random walk model 12 constructs a user electricity consumption behavior sequence network according to the existing historical electricity consumption data, and each historical electricity consumption data is a network node in the electricity consumption behavior sequence network, so that the random walk model 12 takes each network node in the user electricity consumption behavior sequence network as a starting point, respectively performs random walks, generates a random walk sequence corresponding to each walk, and finally outputs each random data in each random walk sequence, and the random data is the random electricity consumption data, realizing the amplification of the first electricity consumption data.

[0070] S102: Iteratively train the control model 13 based on the first electricity consumption data.

[0071] The first electricity consumption data is divided into a training data set and a test data set. The training data set is used to train the control model 13, and the test data set is used to test the pre-established control model 13 and as the input of the control model 13 to extract the electricity consumption behavior characteristics. In this embodiment, the control model 13 is implemented by using a convolutional neural network; specifically, the training data set is input into the control model 13, the electricity consumption behavior characteristics are output, and the electricity consumption behavior characteristics are reversely transmitted into the convolutional neural network to iteratively train the convolutional neural network, optimize the model parameters of the control model 13 until the defined loss function tends to be stable or reaches the maximum number of iterations, and obtain the trained control model 13.

[0072] S103: Extract the electricity consumption behavior characteristics by using the trained control model 13.

[0073] The test data set is used as the first electricity consumption data or directly the first electricity consumption data is input into the trained control model 13, and through the forward transmission of the convolutional neural network in the control model 13, the electricity consumption behavior characteristics corresponding to the first electricity consumption data are output, realizing the extraction of the electricity consumption behavior characteristics.

[0074] S2: Build an electricity consumption simulation platform 2 to simulate the operating conditions of electrical equipment.

[0075] Build a power consumption simulation platform 2 based on the actual power supply network of electrical equipment. The power consumption simulation platform 2 includes loads corresponding one by one to each electrical equipment. The power consumption simulation platform 2 takes the digital control signal formed by integrating the first power consumption data and the corresponding power consumption behavior as the input, and can simulate the operating conditions of the electrical equipment under the control of the digital control signal, and then simulate the simulated power consumption data of the electrical equipment.

[0076] Specifically, the power consumption simulation platform 2 includes an analog circuit 21 and an amplifier circuit 22 electrically connected to the analog circuit 21. The analog circuit 21 includes multiple sets of loads with adjustable parameters. The current and voltage of the loads are reduced by a preset multiple in proportion to those of the electrical equipment, so as to simulate the actual operating conditions of the electrical equipment through the operating conditions of the loads; the amplifier circuit 22 is used to amplify the current and voltage output by the analog circuit 21 by the preset multiple of proportional reduction, so that the simulated power consumption data finally output by the power consumption simulation platform 2 is the same as the actual operating conditions of the electrical equipment, so that subsequent data collection can be effectively carried out.

[0077] In the specific implementation of this embodiment, the parameters of the loads in the analog circuit 21 can also be adjusted or other specifications of loads can be replaced to increase the influence of climate parameters, user switch behavior, and harmonic characteristics on the analog circuit 21 in the analog circuit 21, so that the simulated operating conditions of the analog circuit 21 are closer to the actual operating conditions of the electrical equipment, making the simulated operating conditions of the analog circuit 21 closer to the actual operating conditions of the electrical equipment.

[0078] S3: Apply power theft means to interfere with the simulated power consumption data and collect the second power consumption data.

[0079] [[ID=1,2]]

[0080] S4: Extract the data features of the second power consumption data to obtain a large amount of labeled data.​

[0081] Specifically, data features are extracted from the second electricity consumption data, and labels corresponding to electricity theft means are assigned to the data features, so as to form an association between the data features and the electricity theft means, and a large amount of labeled data carrying electricity theft means information is obtained.

[0082] As Figure 6 shown, step S4 includes the following sub-steps:

[0083] S401: Integrate the second electricity consumption data into a feature matrix.

[0084] Obtain each second electricity consumption data output by the smart meter 31, extract the electricity consumption behavior features of each second electricity consumption data respectively, and integrate the second electricity consumption data and its corresponding electricity consumption behavior features into a feature matrix.

[0085] S402: Calculate the cumulative contribution rate of the principal components by using the principal component analysis method.

[0086] Specifically, first, standardize the feature matrix, and calculate the correlation coefficient matrix according to the standardized feature matrix; then, calculate the eigenvalues in the correlation coefficient matrix, sort the eigenvalues from large to small, and calculate the eigenvectors corresponding to the eigenvalues in turn, and determine the principal components in the electricity consumption behavior features based on the eigenvalues and eigenvectors; finally, calculate the cumulative contribution rate of each principal component in the corresponding electricity consumption behavior features.

[0087] S403: Evaluate the principal components based on the cumulative contribution rate, and extract the data features of the second electricity consumption data.

[0088] Specifically, judge whether the cumulative contribution rate of the principal components is greater than a preset condition. If it is greater than the preset condition (such as 80%), then extract the principal components to form the data features of the second electricity consumption data, otherwise discard them to reduce the influence of some irrelevant interference information on the extracted data features.

[0089] S404: Assign electricity theft means labels to the data features and output a large amount of labeled data.

[0090] Write the electricity theft means information into the corresponding data features, or associate the data features with the electricity theft means, and obtain a large amount of labeled data carrying electricity theft means information, providing data support for realizing electricity theft discrimination and load prediction research.

[0091] The method for generating a large amount of tag data in this embodiment extracts the electricity consumption behavior characteristics by inputting an electricity consumption data including a large amount of historical electricity consumption data and random electricity consumption data into the control model 13, and uses it as a control signal for the electricity consumption simulation platform 2 to control the operating conditions of the load simulation electricity consumption equipment, so as to simulate the actual operating conditions of the electricity consumption equipment; and by applying electricity theft means to collect the simulated electricity consumption data to obtain the second electricity consumption data after being interfered by the electricity theft means, the simulation of the electricity theft means can be realized; finally, by extracting the data characteristics of the second electricity consumption data, a large amount of tag data carrying electricity theft means information can be obtained to make up for the problems of the lack of a large amount of abnormal electricity consumption data, few data samples, and low quality.

[0092] Embodiment 3

[0093] As Figure 7 shown, it is a flowchart of the electricity theft identification method in this embodiment. The electricity theft identification method in this embodiment is implemented based on the method for generating a large amount of tag data in Embodiment 2, and a large amount of tag data is generated to train the electricity theft means model, so as to accurately identify the electricity theft means. Specifically, this embodiment includes the following steps:

[0094] P1: Generate tag data carrying electricity theft means information.

[0095] Specifically, first, extract the electricity consumption behavior characteristics of the user according to the first electricity consumption data, and form a control signal based on the first electricity consumption data and the electricity consumption behavior characteristics to control the electricity consumption simulation platform 2, simulate the operating conditions of the electricity consumption equipment under various electricity consumption behaviors, and output and amplify the generated simulated electricity consumption data; then, intervene in the collection of the simulated electricity consumption data by applying electricity theft means, interfere with the measurement of the simulated electricity consumption data, and respectively output a plurality of second electricity consumption data after being interfered by each electricity theft means; finally, extract the data characteristics of the second electricity consumption data to generate a number of tag data carrying electricity theft means information; for the specific method, refer to the relevant description in Embodiment 2, which will not be elaborated in this embodiment.

[0096] P2: Train the electricity theft identification model 300 based on the tag data.

[0097] Input the label data and the corresponding electricity theft means into a pre-established electricity theft recognition model 300 and perform iterative training on the electricity theft recognition model 300 to obtain a trained electricity theft recognition model 300. In this embodiment, the electricity theft recognition model 300 is implemented using a neural network model; specifically, input the label data and the electricity theft means information into the electricity theft recognition model 300, output the corresponding electricity theft means, and then transmit the electricity theft means back to the electricity theft recognition model 300 for iterative training of the electricity theft recognition model 300 to optimize the model parameters of the electricity theft recognition model 300 until the defined loss function tends to be stable or reaches the maximum number of iterations, obtaining a trained electricity theft recognition model 300.

[0098] P3: Use the trained electricity theft recognition model 300 to identify the electricity theft means.

[0099] Input the electricity consumption data of the user output by the smart meter 31 configured at the user's entry point into the electricity theft recognition model 300. Through the forward transmission of the electricity theft recognition model 300, it is determined whether the smart meter 31 has been subjected to an electricity theft means, and when an electricity theft means has been applied, the category of the electricity theft means is identified, so as to promptly screen and investigate abnormal electricity consumption behaviors.

[0100] The electricity theft recognition method of this embodiment trains the electricity theft recognition model 300 based on the generated massive label data, enabling the trained electricity theft recognition model 300 to accurately identify the category of the electricity theft means, so that the State Grid staff can promptly screen and stop electricity theft behaviors, thereby ensuring the normal and safe operation of the State Grid.

[0101] Embodiment 4

[0102] As Figure 8 shown, it is the control block diagram of the electricity theft recognition system of this embodiment. The electricity theft recognition system of this embodiment can implement the electricity theft recognition method described in Embodiment 3 during operation. Specifically, the electricity theft recognition system of this embodiment includes a massive label data generation platform 100, a model training module 200, and an electricity theft recognition model 300; where:

[0103] The massive label data generation platform 100 generates a number of label data carrying electricity theft means information based on the massive label data generation platform of Embodiment 1. Specifically, the massive label data generation platform 100 controls an electricity consumption simulation platform 2 according to the first electricity consumption data and its corresponding electricity consumption behavior characteristics, so that the smart meter 31 with an electricity theft means applied collects the second electricity consumption data of the electricity consumption simulation platform 2, and extracts the data characteristics of the second electricity consumption data to obtain label data carrying electricity theft means information. For the specific structure of the massive label data generation platform 100, please refer to the relevant description of Embodiment 1, and this embodiment will not be elaborated.

[0104] The model training module 200 is configured to train an electricity theft recognition model 300 according to the label data and the corresponding electricity theft means information. Specifically, the model training module 200 inputs the label data and the corresponding electricity theft means into the electricity theft recognition model 300 and performs iterative training on the electricity theft recognition model 300 to obtain a trained electricity theft recognition model 300.

[0105] The electricity theft recognition model 300 is configured to identify electricity theft means based on the electricity consumption data output by the smart meter 31 configured at the electricity user.

[0106] In the electricity theft recognition system of this embodiment, by setting up a massive label data generation platform, a large amount of label data for training the electricity theft recognition model 300 can be generated, which can improve the training accuracy of the electricity theft recognition model 300, enabling the trained electricity theft recognition model 300 to accurately identify the categories of electricity theft means, so as to promptly distinguish and investigate abnormal electricity consumption behaviors.

Claims

1. A method for generating a large amount of tag data, characterized in that, It includes the following steps: S1: Obtain the first electricity consumption data of the user, and use the first electricity consumption data to train a pre-established control model to extract the electricity consumption behavior characteristics corresponding to the first electricity consumption data; wherein, this step includes the following sub-steps: S101: Obtain the first electricity consumption data of the user, wherein the first electricity consumption data includes existing historical electricity consumption data and random electricity consumption data generated based on the historical electricity consumption data; S102: Divide the first electricity consumption data into a training data set and a test data set, and input the training data set into a pre-established control model to perform iterative training on the control model, optimize the model parameters, and obtain a trained control model; S103: Input the test data set as the first electricity consumption data into the trained control model to extract the electricity consumption behavior characteristics corresponding to the first electricity consumption data; S2: Build an electricity consumption simulation platform, input the first electricity consumption data and its corresponding electricity consumption behavior characteristics into the electricity consumption simulation platform, control the loads in the electricity consumption simulation platform, amplify and output simulated electricity consumption data; S3: Apply at least one electricity theft means to interfere with the metering of the simulated electricity consumption data at the output end of the electricity consumption simulation platform, and output the second electricity consumption data interfered by the electricity theft means; S4: Extract the data characteristics of the second electricity consumption data, associate the data characteristics with the corresponding electricity theft means, and obtain a number of labeled data.

2. The method for generating a large amount of tag data according to claim 1, wherein In step S101, the random electricity consumption data is generated based on the random walk algorithm, and its specific method is: Construct a user electricity consumption behavior sequence network according to the existing historical electricity consumption data, and establish a random walk model based on python. The random walk model takes each network node in the user electricity consumption behavior sequence network as a starting point, respectively performs random walks, generates a random walk sequence corresponding to each walk, and outputs each random data in each random walk sequence. The random data is the random electricity consumption data.

3. The method for generating a large amount of tag data according to claim 1, wherein The electricity consumption behavior characteristics at least include daily electricity consumption, daily maximum / minimum load, average daily load, peak-valley electricity consumption, peak-valley average load, peak-valley difference, night load rate, electricity peak duration, and load curve.

4. The method for generating a large amount of tag data according to claim 1, characterized in that In step S2, the electricity consumption simulation platform includes an analog circuit and an amplifier circuit electrically connected to the analog circuit. The analog circuit includes multiple groups of loads with adjustable parameters. Among them, during the process of simulating the operating conditions, the influence of climate parameters, harmonic characteristics, and user switch behavior on the loads is increased in the analog circuit 21; and the current and voltage of the loads are reduced by a preset multiple in proportion compared with the electrical equipment. The loads are used to simulate electrical equipment; the amplifier circuit is used to amplify the current and voltage at the output end of the analog circuit by the preset multiple in proportion.

5. The method for generating a large amount of tag data according to claim 4, wherein A plurality of smart meters electrically connected to the amplifier circuit are arranged at the output end of the electricity consumption simulation platform. The specific method of step S3 is: Apply a power theft means to each smart meter according to the category of the power theft means, receive analog power consumption data through the smart meter, cause the power theft means to interfere with the measurement of the analog power consumption data, and output second power consumption data generated under different power theft means through each smart meter.

6. The method for generating a large amount of tag data according to claim 1, wherein The step S4 includes the following sub-steps: S401: Obtain the second power consumption data and its corresponding power consumption behavior characteristics to form a feature matrix; S402: Use the principal component analysis method to calculate the principal components of the feature matrix and the cumulative contribution rate of each principal component in the power consumption behavior characteristics; S403: Extract the principal components with a cumulative contribution rate greater than the preset condition according to the cumulative contribution rate of the principal components to form the data characteristics of the second power consumption data; S404: Associate the data characteristics with the corresponding power theft means to obtain a number of tag data carrying power theft means information.

7. A massive tag data generation platform based on the massive tag data generation method according to any one of claims 1-6, characterized in that Include: A control module, configured to extract power consumption behavior characteristics according to the first power consumption data, and generate a control signal based on the first power consumption data and the power consumption behavior characteristics, wherein the first power consumption data includes existing historical power consumption data and random power consumption data generated based on the historical power consumption data; A power consumption simulation platform, configured to simulate the operating conditions of power consumption equipment according to the control signal, output and amplify the analog power consumption data generated by the power consumption simulation platform; A data acquisition and interference module, configured to acquire analog power consumption data and apply a power theft means to interfere with the measurement of the analog power consumption data, and output second power consumption data interfered by the power theft means; A data processing module, configured to extract the data characteristics of the second power consumption data and generate a number of tag data carrying power theft means information.

8. A method for identifying electricity theft, characterized in that, Include the following steps: P1: Generate a number of tag data carrying power theft means information by using the method for generating a large amount of tag data according to any one of claims 1-6; P2: Input the tag data and the corresponding power theft means into a pre-established power theft identification model and perform iterative training on the power theft identification model to obtain a trained power theft identification model; P3: Use the trained power theft identification model to identify the power theft means.

9. An electricity theft identification system, characterized in that, Include: Use the platform for generating a large amount of tag data according to claim 7, configured to generate a number of tag data carrying power theft means information based on a platform for generating a large amount of tag data; A model training module, configured to train a power theft identification model according to the tag data and the corresponding power theft means information; And A power theft identification model, configured to identify the power theft means according to the power consumption data output by the smart meter configured at the power consumption user.

Images