Digital graph fused electricity larceny behavior early warning method, system and device

Abstract

The invention belongs to the field of electric power safety, and particularly relates to an electricity larceny behavior early warning method, system and device based on digital graph fusion. According to the method, a topological graph of each electric power metering node in a transformer area is combined with metering data, and electric power nodes of commercial users and residential users on the topological graph are segmented to obtain a commercial tenant sub-graph and a residential sub-graph. Then, the fluctuation similarity and the increase similarity of power utilization data changes of the power consumers on the two sub-graphs are analyzed in combination with measurement data, and nodes with the high similarity are regarded as safe nodes without the risk of electricity stealing; then, generating each independent risk area according to the residual topological graph, and generating a corresponding feature sequence by combining the power information of each node in the risk area; and finally, utilizing a machine learning algorithm to generate a classification result of the electricity stealing behavior released by each region according to the input feature sequence. According to the invention, the problems of poor precision and insufficient real-time performance of an existing electricity stealing behavior detection method based on big data economy are solved.

Description

Technical Field

[0001] The present invention belongs to the field of electric power safety, and in particular relates to a digital-graphic fusion electricity theft behavior early warning method, system and device. Background Art

[0002] Electricity theft refers to the act of stealing electricity from the power grid by illegal means. On the one hand, electricity theft will cause economic losses to the power company, and on the other hand, it may bring safety hazards to the smooth operation of the power grid. Although there are many ways and types of electricity theft that have been discovered, it can be divided into two categories according to the basic working principle of electricity theft: electricity theft that is unrelated to metering devices and electricity theft that is related to metering devices.

[0003] Among them, electricity theft unrelated to metering devices includes users privately increasing capacity, short-circuiting the inlet and outlet lines of the meter box, and stealing electricity. Although electricity theft through these means is relatively concealed and cannot be discovered during daily inspections, it is possible to quickly locate risk areas based on the data differences of meters at different nodes, and to discover related illegal behaviors through more detailed power inspections. Electricity theft methods related to metering devices include undervoltage theft, undercurrent theft, phase shift theft, differential theft, and so on. The detection and location of this type of electricity theft is relatively difficult, and usually requires manual detection or big data analysis methods based on artificial intelligence to achieve.

[0004] At present, there are two types of electricity theft detection methods widely used in the process of power grid security management: one is the traditional manual detection method, and the other is the method based on big data analysis. The former uses manual sampling to patrol the abnormal electricity consumption caused by electricity theft. However, with the diversification of electricity theft methods, the difficulty of detecting electricity theft by manual detection methods increases, and the probability of finding abnormal electricity users decreases, with a high false alarm rate and missed alarm rate. The latter uses machine learning algorithms such as neural networks to analyze the power information collected by each node in the substation area, and then the big data algorithm identifies and locates the abnormal electricity consumption caused by electricity theft in the substation area based on the substation operation data. The electricity theft detection method based on big data analysis has good generalization ability and can effectively reduce the missed alarm rate. However, the training of the machine learning algorithm in this scheme requires a large amount of abnormal electricity consumption information data of different categories of electricity theft. In the management center of the power grid, the sample data of abnormal information data corresponding to electricity theft behavior is very limited, while the sample data of normal electricity consumption behavior is very large, which makes the model easy to fall into local optimality and slow convergence speed during the training stage. In addition, in the actual application stage, the neural network algorithm needs to process a large amount of power characteristic information generated by all nodes in the substation. In a complex substation including a large number of power users, this will cause the algorithm's parameter scale to be very large, increase the solution's requirements for computing power, and lead to insufficient real-time performance of the solution. Summary of the invention

[0005] In order to solve the problems existing in the existing electricity theft detection methods based on the big data economy, the present invention provides a data-graph fusion electricity theft warning method, system and device.

[0006] The present invention is implemented by the following technical solutions: A digital-graphic fusion electricity theft behavior early warning method comprises the following steps: Obtain the archive information and metering information of all meters in the substation, including the metering data of the day and historical metering data.

[0007] Combined with the archival information, a general topological map is constructed to represent the subordinate relationships among all the gateway meters, measuring switches and electric energy meters in the substation area. According to the type of power user corresponding to each electric energy meter, the general topological map is divided into a merchant submap and a resident submap.

[0008] Based on the load data of the day and the historical metering data, the ICP of the per capita electricity consumption of each residential user, the GRD of the daily average electricity consumption of each commercial user, and the variance of the phase power of each commercial user or residential user are calculated respectively. .

[0009] according to , ICP, and GRD generate the feature vector of each node, calculate the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or merchant subgraph, and merge the nodes with similarity higher than the preset threshold; the electricity meters of the remaining nodes are recorded as independent meters with the risk of electricity theft.

[0010] The metering information of the remaining independent electricity meters and their corresponding measuring switches and gateway meters in the merchant sub-map and the resident sub-map is extracted to generate the feature sequence corresponding to each risk area; and it is input into an electricity theft detection model trained based on MLP to generate the classification result of electricity theft type.

[0011] Based on the classification results output by the electricity theft detection model, the electricity theft behavior of the corresponding power users of each independent electricity meter under the measuring switch with the risk of electricity theft is checked.

[0012] As a further improvement of the present invention, the archive information includes the device identification number, address information, subordinate information, and / or user number of the meter. The subordinate information of the gateway meter includes the device identification number of the measuring switch under its jurisdiction. The subordinate information of the measuring switch includes the device identification number of the upstream gateway meter and the device identification number of the subordinate electric energy meter. The subordinate information of the electric energy meter includes the device identification number of the upstream measuring switch.

[0013] The classification results of the electricity theft types identified by the present invention include: phase shift electricity theft, undervoltage electricity theft, differential expansion electricity theft, undercurrent electricity theft and no electricity theft.

[0014] As a further improvement of the present invention, the method for segmenting the merchant subgraph is as follows: (1) Obtain the overall topology diagram and the user type of the power user corresponding to each electric energy meter contained in it. The overall topology diagram is a tree diagram with each meter as a node and the subordinate relationship between each meter as an edge.

[0015] (2) Delete the nodes corresponding to the electric energy meters belonging to residential users from the overall topology map.

[0016] (3) Delete the measuring switches that no longer have subordinate electric energy meters after deleting the previous step from the overall topology diagram.

[0017] (4) Delete the gateway table that no longer has the range switches under it after the deletion in the previous step from the overall topology map to obtain the merchant sub-map.

[0018] Accordingly, the nodes corresponding to the merchant's electric energy meter are deleted in order, and then the nodes corresponding to the range switch and the gateway meter without subordinate nodes are deleted, and the resident subgraph can be obtained.

[0019] As a further improvement of the present invention, the variance of the phase power of commercial users or residential users is The calculation formula is: , In the above formula, Indicates the i-phase power of the current node at the t-th sampling time of the day; It represents the average power of phase i of the current node on that day; n represents the daily sampling frequency of the meter.

[0020] As a further improvement of the present invention, the calculation formulas for the per capita electricity consumption increase rate ICP of residential users and the daily average electricity consumption growth rate GRD of commercial users are as follows: , In the above formula, and They represent the cumulative electricity volume measured by the current meter on day T and day T-1 respectively; m represents the number of people in the current household of the residential user.

[0021] As a further improvement of the present invention, the calculation method of the fluctuation similarity S1 of any two nodes a and b is as follows: The variance of the three-phase power corresponding to each node , , As the spatial coordinates of the node; Then, the fluctuation similarity S1 between nodes is the Euclidean distance between them, and the calculation formula is as follows: , In the above formula, x 1 ,y 1 and z 1 are the values ​​of the horizontal, vertical and vertical coordinates of node a respectively; x 2 ,y 2 and z 2 are the values ​​of the horizontal, vertical and vertical coordinates of node b respectively.

[0022] As a further improvement of the present invention, the calculation method of the growth similarity S2 of any two nodes a and b is as follows: Encode the ICP or GRP values ​​of nodes a and b in 7 consecutive days into a 7-dimensional feature vector A and B: A=(a1, a2, a3, a4, a5, a6, a7), B=(a1, a2, a3, a4, a5, a6, a7), Then the growth similarity S2 of two nodes is the cosine distance between the corresponding feature vectors, and the calculation formula is: .

[0023] As a further improvement of the present invention, each feature sequence is composed of feature vectors of a measuring switch and its associated gateway table and multiple electric energy meters, and its data format is as follows: , The above formula, D 1 The feature vector representing the gateway table to which the measurement switch belongs; D 2 The characteristic vector representing the measurement switch; It represents the characteristic vectors of the 1st to wth electric energy meters under the range switch.

[0024] The characteristic vector of each meter is composed of part or all of the characteristic values ​​of the following multiple power parameters: Three-phase voltage U A , U B , U C , three-phase current I A ,I B ,I C , secondary current I SC , power factor , line loss value L, line loss conversion rate Δ, number of reverse power abnormalities Ep.

[0025] The present invention also includes a digital-graph fusion electricity theft behavior early warning system, which uses the digital-graph fusion electricity theft behavior early warning method as described above to identify nodes and risk types with electricity theft risks in the production area. The early warning system includes: a topology map generation module, a subgraph segmentation module, a similarity calculation unit, a subgraph merging unit, a feature encoding unit, and a risk prediction unit.

[0026] The topology generation module is used to construct a general topology map for characterizing the subordinate relationship of all meters in the substation based on the archive information of all access meters, measuring switches and electric energy meters in the substation area. The sub-map segmentation module is used to segment the general topology map into business sub-maps and resident sub-maps according to the type of power user corresponding to each electric energy meter.

[0027] The similarity calculation unit is used to calculate the ICP of the per capita electricity consumption of each residential user, the GRD of the daily average electricity consumption of each commercial user, and the variance of the phase power of each commercial user or residential user according to the load data of the day and the historical metering data. ; Then according to , ICP, and GRD generate the feature vector of each node, and calculate the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or merchant subgraph respectively.

[0028] The subgraph merging unit is used to combine the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or the merchant subgraph, and merge the nodes corresponding to any electricity meters in the resident subgraph or the merchant subgraph whose similarity exceeds a preset threshold. The merged nodes are recorded as safe nodes, and the electricity meters of the remaining nodes are recorded as independent meters with the risk of electricity theft.

[0029] The feature encoding unit is used to take the upstream and downstream nodes corresponding to each measuring switch containing an independent meter as a risk area, and generate a feature sequence with the metering information of all meters in the risk area as the feature value. The risk prediction unit includes a power theft detection model, which is used to generate and output a classification result for characterizing whether there is a risk of power theft in the current area based on the input feature sequence.

[0030] The present invention also includes a digital-graph fusion electricity theft behavior warning device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the digital-graph fusion electricity theft behavior warning method as described above is implemented, thereby realizing the identification of areas with possible electricity theft risks and their corresponding electricity theft types based on the topological relationship and metering information of all meters in the substation area.

[0031] The technical solution provided by the present invention has the following beneficial effects: The present invention provides a method for early warning of electricity theft by integrating data and graphs. The method combines the topological graph of each electricity metering node in the substation area with the metering data, and separates the power nodes of commercial users and residential users on the topological graph to obtain two subgraphs. Then, the similarity of the changes in the electricity consumption data of the power users on the two subgraphs is analyzed in combination with the metering data, and the nodes with higher similarity are regarded as safe nodes without the risk of electricity theft. Then, a feature sequence corresponding to the power data of the nodes in each independent risk area in the remaining topological graph is generated. Finally, a machine learning algorithm is used to generate a classification result of the existence of electricity theft in each area based on the input feature sequence.

[0032] The solution provided by the present invention separately identifies the electricity theft risks of different types of electricity users, and performs security pre-analysis and node merging on the nodes in the sub-topology diagrams of various electricity users in combination with actual conditions. This can effectively remove the interference information contained in the feature information of each node in the original substation, thereby greatly reducing the dimension of the input features of the machine learning algorithm, greatly reducing the difficulty of training the network model, improving the dependence of the model's classification accuracy on the sample data scale, and improving the classification accuracy and generalization of the network model.

[0033] In addition, the parameter scale of the network model adopted by the solution of the present invention is smaller, which on the one hand improves the real-time performance of the solution, and on the other hand enables it to be deployed in existing embedded devices (such as smart fusion terminals) in the substation area, thereby improving the practical value of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flowchart of the steps of the method for early warning of electricity theft by integrating data and graphs provided in Example 1 of the present invention.

[0035] Figure 2 This is a typical overall topology diagram of a substation constructed in Example 1 of the present invention.

[0036] Figure 3 This is a flow chart of the steps of the method for segmenting a merchant sub-graph in Embodiment 1 of the present invention.

[0037] Figure 4 This is a flowchart of the steps of the method for segmenting a resident sub-graph in Example 1 of the present invention.

[0038] Figure 5 for Figure 2 The merchant subgraph is divided out of the total topology graph.

[0039] Figure 6 for Figure 2 The resident subgraph is divided out of the overall topological graph.

[0040] Figure 7This is a flowchart of the steps of the method for calculating the fluctuation similarity between any two nodes in Example 1 of the present invention.

[0041] Figure 8 This is a flowchart of the steps of the method for calculating the growth similarity of any two nodes in Example 1 of the present invention.

[0042] Fig. 9 for Figure 5 Distribution map of risk areas included in the merchant submap shown.

[0043] Fig.10 This is a system architecture diagram of the power theft warning system with digital-graphic fusion provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] Example 1 This embodiment provides a method for early warning of electricity theft by integrating data and graphs. The technical idea of ​​this solution is: first, construct a topological graph containing nodes corresponding to all meters in the substation; and use the metering information of each meter as the feature information of the node. Then, according to the type of each electricity user, the overall topological graph of the substation is divided into two subgraphs containing only residential users and only commercial users. Next, analyze the fluctuation and growth of the historical electricity consumption data of each electricity user in the two subgraphs, and then divide the nodes with similar fluctuation and growth into safe nodes, and the remaining nodes into risk nodes that may have electricity theft. Finally, extract the metering data of the electric energy meters of these risk nodes and their upstream measuring switches and gateway meters, and encode them into feature sequences. The input feature sequences are classified by the electricity theft analysis model based on the machine learning algorithm to determine whether there is electricity theft and the type of electricity theft corresponding to the electricity theft.

[0046] Combined with the technical concept of the embodiment described above, it can be seen that, unlike the traditional solution that directly uses the power information obtained by all nodes to detect power theft, the embodiment discusses the two separately according to the different rules of residential power consumption and commercial power consumption, and before inputting the node information into the machine learning algorithm, it predicts whether the node has the risk of power theft based on the changes in the power consumption data of each power user, and then only inputs the power information related to the node with the risk of power theft into the machine learning algorithm. This new data processing strategy can greatly reduce the computational load of the machine learning algorithm and reduce the algorithm's dependence on sample data during the training phase. On the basis of ensuring the final prediction accuracy, it improves the real-time performance of the algorithm and enables the power theft warning solution to be deployed in embedded devices in the substation area.

[0047] Specifically, Figure 1 As shown, the electricity theft behavior early warning method based on data and graph fusion provided in this embodiment includes the following steps: S1: Obtain the archive information and metering information of all meters in the substation. The metering information includes the metering data of the day and the historical metering data.

[0048] In this embodiment, the meters installed in the substation include three types, namely, gateway meters, measuring switches and electric energy meters. Among them, the electric energy meter is installed on the user side to measure the electricity consumption information of each power user. The measuring switch is installed at a branch node of the power supply line and serves as the master meter of several electric energy meters under its jurisdiction. The gateway meter is installed on the main line of the substation and serves as the master meter of several range switches under its jurisdiction.

[0049] For the above three types of meters, the archive information includes the device identification number, address information, subordinate information, and / or user number of the meter. Among them, the device identification number is the exclusive identification code of each meter; the address information corresponds to the installation location of the meter at the node; for the electric energy meter, its archive information also includes the user number, which is a unique account number obtained by each power user when handling the power supply business at the power company. By querying the business database of the power company in combination with the account number, the user type, historical electricity consumption data and member information of each user can also be obtained. Subordinate information refers to the subordinate relationship between each meter in the substation and its upstream or subordinate meters. Specifically, in the topological structure of each meter in the substation, the gateway table is located at the top level, the measuring switch is located at the middle level, and the electric energy meter is located at the bottom level. Therefore, the subordinate information of the gateway table includes the device identification number of the measuring switch under its jurisdiction; the subordinate information of the measuring switch includes the device identification number of the upstream gateway table and the device identification number of the electric energy meter under its jurisdiction; the subordinate information of the electric energy meter includes the device identification number of the upstream measuring switch.

[0050] In this embodiment, the metering information includes a series of power characteristic parameters that can be collected or indirectly calculated between meters, such as: three-phase voltage U A , U B , U C , three-phase current I A ,I B ,I C , secondary current I SC , power factor , line loss value L, line loss conversion rate Δ, number of reverse power abnormalities Ep, and so on.

[0051] S2: According to the three-level pyramid structure of gateway meters, measuring switches and electric energy meters, combined with archival information, a general topological diagram representing the subordinate relationship of all meters in the substation is constructed.

[0052] In this embodiment, the overall topology of the substation area can be generated by the subordinate information of each meter in the substation area. The overall topology is a tree diagram with each meter as a node and the subordinate relationship between each meter as an edge. A typical overall topology includes three layers, the top layer has the least number of gateway meters, followed by the number of measurement switches, and the bottom layer has the most data. Figure 2 As shown, the nodes corresponding to the three types of meters in the overall topology form a typical pyramid structure.

[0053] S3: According to the type of power user corresponding to each power meter, the overall topology map is divided into a business sub-map and a residential sub-map.

[0054] In the scheme implemented in this embodiment, technicians found that the statistical characteristics of electricity consumption data of different types of power users have obvious differences. Therefore, when analyzing the electricity theft behavior of different nodes, different characteristic information should be extracted in combination with different types of power users. Therefore, this embodiment divides different types of users in the substation into different subgraphs, and analyzes the electricity theft behavior of each power user contained in the two subgraphs, which can effectively improve the classification accuracy of the scheme and reduce the parameter scale of the trained machine learning algorithm.

[0055] Specifically, Figure 3 As shown in the figure, the segmentation method of the merchant subgraph is: (1) Obtain the overall topology map and the user type of the power user corresponding to each power meter contained in it.

[0056] (2) Delete the nodes corresponding to the electric energy meters belonging to residential users from the overall topology map.

[0057] (3) Delete the measuring switches that no longer have subordinate electric energy meters after deleting the previous step from the overall topology diagram.

[0058] (4) Delete the gateway table that no longer has the range switches under it after the deletion in the previous step from the overall topology map to obtain the merchant sub-map.

[0059] Correspondingly, such as Figure 4 As shown in Figure 2, the segmentation method of the resident subgraph is: (1) Obtain the overall topology map and the user type of the power user corresponding to each power meter contained in it.

[0060] (2) Delete the nodes corresponding to the electricity meters belonging to commercial users from the overall topology map.

[0061] (3) Delete the measuring switches that no longer have subordinate electric energy meters after deleting the previous step from the overall topology diagram.

[0062] (4) Delete the gateway table that no longer has the range switches under it after the deletion in the previous step from the overall topology map to obtain the merchant sub-map.

[0063] For example, based on the segmentation method of the business subgraph and the resident subgraph introduced above, Figure 2 The typical overall topology shown in the figure can be segmented to obtain Figure 5 The merchant sub-graph shown and Figure 6 The resident subgraph is shown.

[0064] S4: Based on the load data of the day and the historical metering data, calculate the ICP of the per capita electricity consumption of each residential user, the GRD of the daily average electricity consumption of each commercial user, and the variance of the phase power of each commercial user or residential user. .

[0065] Considering that in real substation scenarios, each commercial subgraph or residential subgraph still contains a large number of power user nodes, and most of the power user nodes do not actually steal electricity, this embodiment combines the residential subgraph and the commercial subgraph to exclude the security nodes where there is obviously no power theft.

[0066] In the process of designing the criteria for excluding safety nodes, the technical personnel in this case found that for ordinary residential users, the changes in their daily electricity consumption will be affected by natural climate and daily activities, and will show obvious statistical regularities in the time domain. For example, during the high temperatures in summer and low temperatures in winter, due to the increase in cooling and heating demand, the electricity consumption of ordinary residential users will increase significantly compared to other periods. In addition, the electricity consumption data of residential users on weekends is relatively stable, while on weekdays, the electricity consumption of residential users will increase significantly after family members get off work or school, etc.

[0067] Theoretically, combining the time domain changes of the user's electricity consumption over a long period of time will help analyze whether there are abnormalities in the user's electricity consumption data. However, this data analysis strategy has a large amount of data processing and is not suitable for practical application. In response to this situation, the technicians of this embodiment have taken a different approach and considered first analyzing the fluctuations and growth of the electricity consumption data of users of the same type, and then analyzing the similarity of the fluctuations and growth of electricity consumption of power users at different nodes in the same period, and finally combining the similarity of the changes in electricity consumption data between nodes to determine whether the corresponding node has a risk of electricity theft.

[0068] Taking residential users as an example, if the fluctuation and growth of electricity consumption of any two users in a certain period are similar, it means that the changes in their electricity consumption data are in line with common sense, and the changes in electricity consumption may be caused by the local natural climate or group activity patterns. At this time, both nodes can be considered safe nodes. On the contrary, when the fluctuation and growth of electricity consumption of any two users in a certain period are very different, it means that the changes in their electricity consumption data may not be in line with common sense. This data difference may be due to the inconsistency between the metered electricity consumption and the actual electricity consumption of one of the two, that is, the existence of electricity theft. At this time, these two nodes should be considered as nodes with the risk of electricity theft.

[0069] Correspondingly, for commercial users, the business activities of various businesses in the same area are usually similar; and the business activities are often positively correlated with the electricity consumption of the businesses. Therefore, this embodiment first analyzes the fluctuation and growth of the electricity consumption data of each commercial user, and then analyzes whether the fluctuation and growth of electricity consumption of power users at different nodes in the same period are similar. If so, it is determined as a safe node, otherwise it is regarded as a risky node.

[0070] Specifically, in this embodiment, the variance of the phase power is set The ICP of per capita electricity consumption is used as an indicator to evaluate the fluctuation and growth of daily electricity consumption of residential users, and the variance of phase power is used as the The average daily electricity consumption growth rate GRD is used as an indicator to evaluate the fluctuation and growth of daily electricity consumption of commercial users.

[0071] Among them, the variance of the phase power of commercial users and residential users is The same formula is used for calculation, which is as follows: , In the above formula, Indicates the i-phase power of the current node at the t-th sampling time of the day; It represents the average power of phase i of the current node on that day; n represents the daily sampling frequency of the meter.

[0072] Among them, the ICP of per capita electricity consumption of residential users refers to the difference between the per capita electricity consumption of residential users on the current day and the per capita electricity consumption of the previous day. The GRD of daily average electricity consumption growth rate of commercial users refers to the average of the daily growth rate of electricity consumption of commercial users over a consecutive 30 days compared with the same period last year. The calculation formulas of ICP and GRD are as follows: , In the above formula, and They represent the cumulative electricity volume measured by the current meter on day T and day T-1 respectively; m represents the number of people in the current household of the residential user.

[0073] S5: According to , ICP, and GRD generate the feature vector of each node, calculate the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or merchant subgraph, and merge the nodes with similarity higher than the preset threshold; the electricity meters of the remaining nodes are recorded as independent meters with the risk of electricity theft.

[0074] In this embodiment, if Figure 7 As shown, the calculation method of the fluctuation similarity S1 of any two nodes a and b is as follows: The variance of the three-phase power corresponding to each node , , As the spatial coordinates of the node, we have: The fluctuation similarity S1 between nodes is the Euclidean distance between them, and the calculation formula is as follows: , In the above formula, x 1 ,y 1 and z 1 are the values ​​of the horizontal, vertical and vertical coordinates of node a respectively; x 2 ,y 2 and z 2 are the values ​​of the horizontal, vertical and vertical coordinates of node b respectively.

[0075] like Figure 8 As shown, the calculation method of the growth similarity S2 of any two nodes a and b is as follows: Encode the ICP or GRP values ​​of nodes a and b in 7 consecutive days into a 7-dimensional feature vector A and B: A=(a1, a2, a3, a4, a5, a6, a7), B=(a1, a2, a3, a4, a5, a6, a7), Then the growth similarity S2 of two nodes is the cosine distance between the corresponding feature vectors, and the calculation formula is: .

[0076] In this embodiment, the fluctuation similarity S1 and growth similarity S2 of any two electric energy meters in the resident subgraph or the merchant subgraph are first calculated. Next, the nodes corresponding to each electric energy meter are first merged according to S1, and then the nodes corresponding to each electric energy meter are merged in the second round according to S2. The operation strategy of the two rounds of merging is: First, each node whose S1 or S2 is greater than or equal to the preset similarity threshold is classified into the safe node set, and then each node whose S1 or S2 is less than the preset similarity threshold is classified into the risk node set. Finally, considering that in the calculated similarity results of any two nodes, some nodes have a high similarity with one of the nodes (safe nodes) and a low similarity with other nodes (risk nodes), these nodes should be regarded as safe nodes according to the actual situation. That is, the nodes that appear in both the safe node set and the risk node set are removed from the risk node set.

[0077] S6: Extract the metering information of the remaining independent electricity meters and their corresponding measuring switches and gateway meters in the merchant sub-map and the resident sub-map, and generate the feature sequence corresponding to each risk area. Input each feature sequence into an electricity theft detection model trained based on MLP to generate the classification result of electricity theft type.

[0078] After the two rounds of merging of safe nodes in step S5, only a small number of risk nodes remain in the resident subgraph or the merchant subgraph. Fig. 9 As shown, this step defines the sub-graph area consisting of the same measuring switch and its associated gateway meter and multiple independent meters contained in the resident sub-graph or the merchant sub-graph as a risk area, and each resident sub-graph or merchant sub-graph may contain multiple risk areas. Next, this embodiment extracts the metering information of all meters in each risk area and encodes the metering information into a feature sequence.

[0079] Specifically, each feature sequence is composed of feature vectors of a measurement switch and its associated gateway table and multiple electric energy meters, and its data format is as follows: ,

[0080] In the above formula, D 1 The feature vector representing the gateway table to which the measurement switch belongs; D 2 The characteristic vector representing the measurement switch; It represents the characteristic vectors of the 1st to wth electric energy meters under the range switch.

[0081] The characteristic vector of each meter is composed of a series of characteristic values ​​of power parameters, and the characteristic vector of each meter contains parameters selected from the following power characteristic parameters: Three-phase voltage U A , U B , U C , three-phase current I A ,I B ,I C , secondary current I SC , power factor , line loss value L, line loss conversion rate Δ, number of reverse power abnormalities Ep.

[0082] It should be additionally explained that, for any meter, its characteristic vector may select all of the above-mentioned power characteristic parameters at the same time, or may select a part of the above-mentioned power characteristic parameters.

[0083] The generated feature sequences corresponding to each risk area will be input into a pre-trained electricity theft detection model, which is used to generate the corresponding classification results of the electricity theft type according to the input feature sequence. In this embodiment, the classification results of the electricity theft types identified by the model include five types, namely: phase shift electricity theft, undervoltage electricity theft, differential electricity theft, undercurrent electricity theft, and no electricity theft. Among them, "no electricity theft" means that the risk of electricity theft does not exist in the risk area predicted by the model.

[0084] In the more optimized solution of this embodiment, in order to eliminate the interference of the dimensions of different power characteristic parameters on the prediction performance of the electricity theft detection model, it is necessary to normalize the value of each characteristic parameter contained in the generated characteristic sequence. The calculation formula of the normalization process is as follows: , In the above formula, x Represents the original value of the feature parameter; Represents the normalized result of the feature parameters; and Represent the maximum and minimum values ​​of the characteristic parameters respectively.

[0085] In addition, it should be additionally explained that: the solution provided in this embodiment can use the same trained electricity theft detection model to classify the feature sequences in the two sub-maps at the same time, or use a large number of feature sequences from risk areas in the commercial sub-map and the residential sub-map to train a corresponding electricity theft detection module respectively, and then use the two electricity theft detection modules to perform electricity theft risk analysis on feature sequences from different sources.

[0086] S8: Generate warning information based on the classification results output by the electricity theft detection model, and conduct electricity theft investigation on the electricity users corresponding to the independent electricity meters under the measuring switches with electricity theft risks.

[0087] In the scheme of this embodiment, the classification result output by the electricity theft detection model in the previous step can reflect whether there is electricity theft in the risk area and the type of corresponding electricity theft. In this step, the management center of the power grid can generate early warning information based on the corresponding classification results, and send work orders to the operation and maintenance personnel in the area. The relevant personnel will arrive at the phenomenon to manually check the meters of each power user, find the object of electricity theft, consolidate the evidence and initiate subsequent legal procedures to deal with the illegal behavior.

[0088] Example 2 On the basis of the scheme in Example 1, Figure 2 As shown, this embodiment further provides a digital-graph fusion electricity theft behavior warning system, which is a computer program for implementing the solution of embodiment 1, and can be stored in a storage medium or a computer device. Then, when the computer program is running, the digital-graph fusion electricity theft behavior warning method as in embodiment 1 is implemented, and the nodes and risk types with electricity theft risks in the substation area are identified based on the monitoring data generated in real time in the substation area. Fig.10 As shown, the early warning system includes: a topological map generation module, a sub-map segmentation module, a similarity calculation unit, a sub-map merging unit, a feature encoding unit, and a risk prediction unit.

[0089] The topology generation module is used to construct a general topology map for characterizing the subordinate relationship of all meters in the substation based on the archive information of all access meters, measuring switches and electric energy meters in the substation area. The sub-map segmentation module is used to segment the general topology map into business sub-maps and resident sub-maps according to the type of power user corresponding to each electric energy meter.

[0090] The similarity calculation unit is used to calculate the ICP of the per capita electricity consumption of each residential user, the GRD of the daily average electricity consumption of each commercial user, and the variance of the phase power of each commercial user or residential user according to the load data of the day and the historical metering data. ; Then according to , ICP, and GRD generate the feature vector of each node, and calculate the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or merchant subgraph respectively.

[0091] The subgraph merging unit is used to combine the fluctuation similarity S1 and growth similarity S2 of any two electricity meters in the resident subgraph or the merchant subgraph, and merge the nodes corresponding to any electricity meters in the resident subgraph or the merchant subgraph whose similarity exceeds a preset threshold. The merged nodes are recorded as safe nodes, and the electricity meters of the remaining nodes are recorded as independent meters with the risk of electricity theft.

[0092] The feature encoding unit is used to take the upstream and downstream nodes corresponding to each measuring switch containing an independent meter as a risk area, and generate a feature sequence with the metering information of all meters in the risk area as the feature value. The risk prediction unit includes a power theft detection model, which is used to generate and output a classification result for characterizing whether there is a risk of power theft in the current area based on the input feature sequence.

[0093] Example 3 This embodiment provides a digital-graph fusion electricity theft behavior warning device, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the digital-graph fusion electricity theft behavior warning method as in Example 1 is implemented, thereby realizing the identification of areas with possible electricity theft risks and their corresponding electricity theft types based on the topological relationship and metering information of all meters in the substation area.

[0094] The digital-graphic fusion electricity theft behavior warning device provided in this embodiment is essentially a computer device. In actual application, the computer device can be an embedded device and deployed in a concentrator or fusion terminal with strong data processing capabilities in the station. It can also be a sample of independent computer devices, such as laptops, tablet computers, desktop computers, or medium and large computer devices such as rack servers, blade servers, tower servers, or cabinet servers (including independent servers, or server clusters composed of multiple servers) that can execute computer programs.

[0095] The computer device of this embodiment at least includes but is not limited to: a memory and a processor that can be connected to each other through a system bus. In this embodiment, the memory (i.e., a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a random access memory (RAM), a static random access memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory may be an internal storage unit of a computer device, such as a hard disk or a memory of the computer device. In other embodiments, the memory may also be an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device. Of course, the memory may also include both an internal storage unit of a computer device and an external storage device thereof. In this embodiment, the memory is generally used to store an operating system and various application software installed on the computer device. In addition, the memory may also be used to temporarily store various types of data that have been output or are to be output.

[0096] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of a computer device.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A digital-graphic fusion electricity theft behavior early warning method, characterized in that: It includes the following steps: Obtain archive information and metering information of all meters in the substation area, including daily metering data and historical metering data; Combined with archival information, a general topological diagram is constructed to represent the subordinate relationships among all gateway meters, measuring switches and electric energy meters in the substation area; and dividing the total topology map into a business sub-map and a resident sub-map according to the type of power user corresponding to each power meter; Based on the load data of the day and the historical metering data, the ICP of the per capita electricity consumption of each residential user, the GRD of the daily average electricity consumption of each commercial user, and the variance of the phase power of each commercial user or residential user are calculated respectively. ; according to , ICP, GRD generate the feature vector of each node, calculate the fluctuation similarity S1 and growth similarity S2 of any two electric energy meters in the resident subgraph or merchant subgraph, and merge the nodes with similarity higher than the preset threshold; the electric energy meters of the remaining nodes are recorded as independent electric energy meters with the risk of electricity theft; Extracting metering information of each remaining independent electric meter and its corresponding measuring switch and gateway meter in the merchant sub-map and the resident sub-map, and generating a feature sequence corresponding to each risk area; And input it into an electricity theft detection model trained based on MLP to generate a classification result of electricity theft type; Based on the classification result output by the electricity theft detection model, the electricity theft behavior of the corresponding power users of each independent electricity meter under the measuring switch with the risk of electricity theft is checked.

2. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 1, characterized in that: The file information includes the device identification number, address information, subordinate information, and / or user number of the meter; the subordinate information of the gateway meter includes the device identification number of the measuring switch under it; the subordinate information of the measuring switch includes the device identification number of the upstream gateway meter and the device identification number of the electric energy meter under it; the subordinate information of the electric energy meter includes the device identification number of the upstream measuring switch; The classification results of the electricity theft types include: phase shift electricity theft, undervoltage electricity theft, differential expansion electricity theft, undercurrent electricity theft and no electricity theft.

3. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 2, characterized in that: The method for segmenting the merchant subgraph is as follows: (1) Obtaining the overall topology map and the user type of the power user corresponding to each electric energy meter contained therein; The overall topology diagram is a tree diagram with each meter as a node and the subordinate relationship between each meter as an edge; (2) deleting the nodes corresponding to the electric energy meters belonging to the residential users from the overall topology map; (3) Deleting the measuring switches that no longer have subordinate electric energy meters after the above step is deleted from the overall topology diagram; (4) Deleting the gateway table that no longer has the range switches under it after the deletion in the previous step from the overall topology map, thereby obtaining the merchant sub-map; Accordingly, the nodes corresponding to the electricity meters of the merchants are deleted first, and then the nodes corresponding to the range switches and gateway meters without subordinate nodes are deleted, so as to obtain the resident subgraph.

4. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 3, characterized in that: The variance of the phase power of commercial or residential users The calculation formula is: ， In the above formula, Indicates the i-phase power of the current node at the t-th sampling time of the day; It represents the average power of phase i of the current node on that day; n represents the daily sampling frequency of the meter.

5. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 4, characterized in that: The calculation formula for the ICP of the per capita electricity consumption of residential users and the GRD of the average daily electricity consumption of commercial users is as follows: ， In the above formula, and They represent the cumulative electricity volume measured by the current meter on day T and day T-1 respectively; m represents the number of people in the current household of the residential user.

6. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 5, characterized in that: The calculation method of the fluctuation similarity S1 between any two nodes a and b is as follows: The variance of the three-phase power corresponding to each node , , As the spatial coordinates of the node; Then, the fluctuation similarity S1 between nodes is the Euclidean distance between them, and the calculation formula is as follows: ， In the above formula, x 1 ,y 1 and z 1 are the values ​​of the horizontal, vertical and vertical coordinates of node a respectively; x 2 ,y 2 and z 2 are the values ​​of the horizontal, vertical and vertical coordinates of node b respectively.

7. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 1, characterized in that: The calculation method of the growth similarity S2 of any two nodes a and b is as follows: Encode the ICP or GRP values ​​of nodes a and b in 7 consecutive days into a 7-dimensional feature vector A and B: A=(a1, a2, a3, a4, a5, a6, a7), B=(a1, a2, a3, a4, a5, a6, a7), Then the growth similarity S2 of two nodes is the cosine distance between the corresponding feature vectors, and the calculation formula is: 。 8. The method for early warning of electricity theft based on digital-graphic fusion as claimed in claim 7, characterized in that: Each feature sequence is composed of the feature vectors of the measurement switch and its associated gateway table and multiple electric energy meters, and its data format is as follows: ， The above formula, D 1 The characteristic vector representing the gateway table to which the measurement switch belongs; i>D 2 The characteristic vector representing the measurement switch; Then it represents the characteristic vector of the 1st to wth electric energy meters under the range switch; The characteristic vector of each meter is composed of part or all of the characteristic values ​​of the following multiple power parameters: Three-phase voltage U A , U B , U C , three-phase current I A ,I B ,I C , secondary current I SC , power factor , line loss value L, line loss conversion rate Δ, number of reverse power abnormalities Ep.

9. A digital-graphic fusion electricity theft warning system, characterized in that: It adopts the electricity theft behavior early warning method of digital-graph fusion as described in any one of claims 1 to 8 to identify nodes and risk types with electricity theft risks in the test area; The electricity theft warning system comprises: A topology map generation module is used to construct a general topology map for representing the subordinate relationship of all meters in the substation area based on the archive information of all the gateway meters, measuring switches and electric energy meters in the substation area; A subgraph segmentation module, which is used to segment the general topology graph into a merchant subgraph and a resident subgraph according to the type of power user corresponding to each electric energy meter; The similarity calculation unit is used to calculate the per capita electricity consumption increase ICP of each residential user, the daily average electricity consumption growth rate GRD of each commercial user, and the variance of the phase power of each commercial user or residential user based on the load data of the day and the historical metering data. ; Then according to , ICP, GRD generate the feature vector of each node, and calculate the fluctuation similarity S1 and growth similarity S2 of any two electric energy meters in the resident subgraph or merchant subgraph respectively; A subgraph merging unit is used to combine the fluctuation similarity S1 and growth similarity S2 of any two electric energy meters in the resident subgraph or the merchant subgraph, merge the nodes corresponding to any electric energy meters in the resident subgraph or the merchant subgraph whose similarity exceeds a preset threshold, and record the merged nodes as safe nodes, and record the electric energy meters of the remaining nodes as independent electric energy meters with the risk of electricity theft; A feature encoding unit, which is used to take the upstream and downstream nodes corresponding to each measuring switch containing an independent electric meter as a risk area, and generate a feature sequence with the metering information of all meters in the risk area as feature values; The risk prediction unit includes an electricity theft detection model, which is used to generate and output a classification result for characterizing whether there is an electricity theft risk in the current area based on an input feature sequence.

10. A digital-graphic fusion electricity theft warning device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, it implements the method for early warning of electricity theft by integrating data and graph as described in any one of claims 1 to 8, thereby identifying areas where there may be risks of electricity theft and their corresponding types of electricity theft based on the topological relationship and metering information of all meters in the substation.

Images