Anti-electricity theft knowledge graph optimization method and system based on dynamic update of incremental data

By adopting the knowledge graph optimization method of dynamic update of incremental data in anti-power detection, combined with multi-factor typical correlation analysis and block incremental learning algorithm, the problems of low accuracy and efficiency of power detection in the existing technology are solved, and high efficiency is achieved Identification and traceability analysis of electricity theft behavior.

CN114330416BActive Publication Date: 2024-11-05STATE GRID SHANDONG ELECTRIC POWER CO MARKETING SERVICE CENT (MEASURING CENT)
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202111424886.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-11-26
Publication Date
2024-11-05
Estimated Expiration
2041-11-26

AI Technical Summary

Technical Problem

The existing anti-powered detection methods are difficult to identify hidden relationships in massive electricity consumption data, resulting in poor accuracy of powered detection, especially when processing high-dimensional data and dynamic update data, and there is difficulty in determining powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered powered power A problem of insufficient.

Method used

The anti-powered power plagiarism knowledge graph optimization method based on dynamic update of incremental data is adopted to integrate the knowledge graph containing prior knowledge with machine learning algorithms, and the anti-powered power plagiarism model is updated through multi-factor typical correlation analysis and block incremental learning algorithm. Build an optimized anti-powered knowledge graph to realize fine-grained correlation analysis and identification of powered powered behavior.

Benefits of technology

It improves the accuracy of identifying power theft behavior and the efficiency of anti-power inspection, shortens data analysis time, saves computing resources, and provides traceability data support for power theft behavior.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114330416B_ABST
    Figure CN114330416B_ABST
Patent Text Reader

Abstract

The present invention provides an anti-electricity theft knowledge graph optimization method and system based on dynamic update of incremental data, which obtains newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types; performs multi-factor typical correlation analysis based on the newly added collection data to obtain newly added electricity consumption behavior features, and adopts a block incremental learning algorithm to update the feature correlation coefficient value of the typical correlation analysis; based on the obtained newly added electricity consumption behavior features and the updated typical correlation analysis correlation coefficient value, a logistic regression algorithm is adopted to construct a new anti-electricity theft model class based on the newly added electricity theft type, and the parameters of the original anti-electricity theft model are updated and optimized; based on various types of anti-electricity theft models newly added and updated based on the electricity theft type, a bottom-level subgraph and a top-level subgraph are constructed, and subgraph matching is adopted to fuse and process the anti-electricity theft knowledge in the anti-electricity theft knowledge graph to obtain an optimized anti-electricity theft knowledge graph; the present invention improves the efficiency of electricity theft behavior identification and anti-electricity theft inspection.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of smart electricity consumption technology, and in particular to an anti-electricity theft knowledge graph optimization method and system based on dynamic updating of incremental data. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] With the development and expansion of the ubiquitous power Internet of Things (IoT), various monitoring, inspection, and measurement technologies within power grids can measure and monitor electricity usage in real time or near real time. This provides large-scale, high-dimensional, multi-source, and highly timely data for identifying electricity theft, accumulating methods, and investigating theft. This data supports the identification and source tracing of theft. Furthermore, with the advancement of market-oriented reforms in the power industry, power companies are gradually being put on the same footing as other businesses. In power theft investigations, if power management departments fail to allocate adequate administrative enforcement resources and monitor insufficient monitoring evidence, there is a risk of overstepping their authority in investigating and handling theft, potentially leading to serious legal consequences and public opinion crises for power companies. To achieve energy cost recovery and safe and stable grid operation, updating anti-theft identification models and related knowledge based on time-shifted data is crucial for the operation and development of power systems.

[0004] The inventors discovered that current anti-electricity theft detection methods primarily include state detection and manual investigation. State detection is primarily based on historical sample data and employs machine learning algorithms to train classifiers, thereby enabling the identification of electricity theft. However, before training the classifier, feature selection must be performed on high-dimensional data. Manual screening is tedious and time-consuming, and existing methods all focus on one-dimensional electricity usage data. However, user electricity usage is affected by many uncontrollable factors, such as season, lifestyle, and weather conditions. It is often difficult to identify periodic or aperiodic patterns in electricity usage with one-dimensional electricity usage data. Furthermore, human effort is unable to identify hidden correlations within massive amounts of electricity usage data. The inadequacy of feature data correlation analysis often leads to poor electricity theft detection accuracy. With the development of relevant monitoring technologies, various types of electricity usage data are being recorded in greater detail, and as time passes, various types of collected user data are constantly updated. To improve the accuracy of anti-electricity theft detection, dynamically updated, fine-grained correlation analysis is essential. However, during anti-electricity theft inspections, identifying electricity theft is difficult and evidence is insufficient. This is particularly true for temporary theft, as staff are inefficient in detecting theft during surprise inspections. Summary of the Invention

[0005] In order to address the shortcomings of the existing technology, the present invention provides an anti-electricity theft knowledge graph optimization method and system based on dynamic updating of incremental data, which integrates the knowledge graph containing prior knowledge with the machine learning algorithm, conducts fine-grained correlation mining based on massive electricity consumption collection data, continuously and dynamically updates and improves various anti-electricity theft models, and uses the knowledge graph to realize effective and intuitive electricity theft reasoning knowledge tracing, thereby improving the efficiency of electricity theft behavior identification and anti-electricity theft inspection work.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] The first aspect of the present invention provides an anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data.

[0008] An anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data, which obtains newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types;

[0009] Preprocess the acquired data;

[0010] Based on the newly collected data, a multi-factor canonical correlation analysis is performed to obtain the new electricity consumption behavior characteristics, and the block incremental learning algorithm is used to update the characteristic correlation coefficient value of the canonical correlation analysis;

[0011] Based on the newly acquired electricity consumption behavior characteristics and the updated canonical correlation analysis correlation coefficient values, a logistic regression algorithm is used to construct a new anti-electricity theft model class based on the newly added electricity theft types, and the parameters of the original anti-electricity theft model are updated and optimized;

[0012] Based on the newly added and updated anti-electricity theft models of various types of electricity theft, the bottom subgraph and the top subgraph are constructed. The anti-electricity theft knowledge in the anti-electricity theft knowledge graph is fused and processed using subgraph matching to obtain the optimized anti-electricity theft knowledge graph.

[0013] Furthermore, the preprocessing includes: data cleaning, data deletion, data completion and normalization.

[0014] Furthermore, a multi-factor canonical correlation analysis is performed based on the newly collected data to obtain new electricity consumption behavior characteristics. The block incremental learning algorithm is used to update the characteristic correlation coefficient value of the canonical correlation analysis, including:

[0015] Based on the newly acquired data of the newly acquired features, the multi-correlation factor canonical correlation analysis method is used to calculate the overall linear relationship between the sample feature sets, and the canonical correlation is obtained as the new electricity consumption behavior feature of the anti-electricity theft model;

[0016] Divide the collected data of various new and old collection features into incremental data blocks of the same size according to the same time window;

[0017] Using the projection weight matrix of the incremental data block and the previous batch of incremental data blocks, the projection vector is iteratively updated to obtain the projection weight matrix of the newly added incremental data blocks in the current batch;

[0018] Based on the projection weight matrix of the incremental data blocks newly added in the current batch, the maximum correlation coefficient of the newly added batch collected data is obtained, and the maximum correlation coefficient between the newly added features and the updated features is obtained.

[0019] Furthermore, based on the newly acquired electricity consumption behavior characteristics and the updated canonical correlation analysis correlation coefficient values, a logistic regression algorithm is used to construct a new anti-electricity theft model class based on the newly added electricity theft types, and the parameters of the original anti-electricity theft model are updated and optimized, including:

[0020] Data definition is performed based on the newly acquired electricity consumption behavior characteristics and the maximum correlation coefficient of each characteristic of the newly acquired batch data;

[0021] Initialize the anti-electricity theft model parameters for the newly added electricity theft categories based on the newly added electricity theft types, and initialize the parameters of various anti-electricity theft models built based on the logistic regression algorithm;

[0022] Based on the newly added feature data and the updated maximum correlation coefficient, an iterative algorithm is used to calculate and update the model parameters;

[0023] Substitute the updated model parameters into the objective function to determine whether the error between the objective function and the actual electricity theft identification is less than the error threshold. If yes, proceed to the next step; otherwise, return to the previous step.

[0024] The objective function parameters are fixed and the various anti-electricity theft models updated this time are output.

[0025] Furthermore, various anti-electricity theft models constructed based on the logistic regression algorithm are single-category anti-electricity theft models with different input features and different parameter values ​​constructed according to different electricity theft categories.

[0026] Furthermore, based on the newly added and updated anti-electricity theft models of various types of electricity theft, bottom-level subgraphs and top-level subgraphs are constructed. Subgraph matching is used to fuse and process the anti-electricity theft knowledge in the anti-electricity theft knowledge graph to obtain an optimized anti-electricity theft knowledge graph, including:

[0027] Define the electricity theft category and anti-electricity theft knowledge graph based on historical records, and divide the anti-electricity theft knowledge graph into the bottom subgraph and the top subgraph;

[0028] In the process of constructing the underlying subgraph, entities and attributes are extracted based on the historically accumulated anti-electricity theft record information, and triples and multitudes are defined within the knowledge graph;

[0029] Based on expert experience, the acquired multiple factors, and the corresponding maximum correlation coefficients, anti-electricity theft triples and multitudes are constructed, and the associated trigger constraints and suspicious attributes are output as the input of the top-level subgraph.

[0030] The top-level subgraph is constructed using multi-tuples. Based on the anti-theft model constructed based on different electricity theft categories, the suspicious attributes are used as the anti-theft model input, and the electricity theft category is used as the model output. Different anti-theft models are obtained for different electricity theft categories. The characteristic parameter values ​​of each anti-theft model are extracted as the associated attribute values. The suspicious attributes are used as the head entity of the top-level subgraph, and the electricity theft category is used as the tail entity of the top-level subgraph. The output of the top-level subgraph is the electricity theft category.

[0031] Based on the acquired new features and the updated maximum correlation coefficient value, the new features are added to each anti-electricity theft model as feature input, and the multi-factor correlation virtual attribute feature value is updated based on the updated maximum correlation coefficient value, and each anti-electricity theft model is updated, and then the feature parameters of each model are updated. Based on the model parameter value, the features of each model are added and eliminated, and the associated information of the knowledge graph is updated according to the updated feature parameters of each model.

[0032] Furthermore, the features of each model are added or eliminated, including:

[0033] Add or remove new features, and retain or delete existing features.

[0034] The second aspect of the present invention provides an anti-electricity theft knowledge graph optimization system based on dynamic updating of incremental data.

[0035] An anti-electricity theft knowledge graph optimization system based on dynamic update of incremental data, including:

[0036] The data acquisition module is configured to: acquire newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types;

[0037] The preprocessing module is configured to: preprocess the acquired data;

[0038] The correlation calculation module is configured to: perform a multi-factor canonical correlation analysis based on the newly collected data to obtain new electricity consumption behavior characteristics, and use a block incremental learning algorithm to update the characteristic correlation coefficient value of the canonical correlation analysis;

[0039] The model building and optimization module is configured to: based on the newly acquired electricity consumption behavior characteristics and the updated canonical correlation analysis correlation coefficient value, use a logistic regression algorithm to build a new anti-electricity theft model class based on the newly added electricity theft type, and update and optimize the parameters of the original anti-electricity theft model;

[0040] The knowledge graph optimization module is configured to: construct bottom-level subgraphs and top-level subgraphs based on various new and updated anti-electricity theft models of electricity theft types, use subgraph matching to fuse and process the anti-electricity theft knowledge in the anti-electricity theft knowledge graph, and obtain an optimized anti-electricity theft knowledge graph.

[0041] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in the first aspect of the present invention.

[0042] The fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in the first aspect of the present invention are implemented.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention uses canonical correlation analysis to mine fine-grained correlations between multivariate data sets. On the one hand, it achieves feature selection of the overall correlation of high-dimensional data, providing periodic electricity consumption variation characteristics for anti-electricity theft identification; on the other hand, it supports the reconstruction of the anti-electricity theft model based on the characteristics of newly added electricity theft factors, further improving the accuracy of electricity theft prediction.

[0045] 2. The present invention uses block incremental learning over time to update the correlation information between features, and updates the feature data sets of different anti-electricity theft models based on the electricity theft category. This not only improves the accuracy of electricity theft behavior identification, but also shortens the running time of data analysis and mining through the integration of basic and value-added methods when processing large amounts of data, thereby saving storage space and computing resources.

[0046] 3. The present invention uses a knowledge graph to store and display the identification and reasoning process of electricity theft that integrates expert prior knowledge and machine learning, mines the relationship between collected data features, calculates the correlation between multi-factor features, expands anti-electricity theft features containing complex implicit relationships, and constructs the underlying subgraph of the knowledge graph based on this, thereby providing basic traceability data support for electricity theft.

[0047] 4. The present invention uses a logistic regression algorithm to construct anti-electricity theft models for different types of electricity theft, and intuitively and effectively connects and displays information between machine learning processing and reasoning processes, thereby realizing traceability analysis based on the electricity theft behavior identification relationship network, and thus providing a verification basis for the implementation process of anti-electricity theft work.

[0048] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0050] Figure 1 This is the overall flow chart of the anti-electricity theft knowledge graph optimization provided in Example 1 of the present invention.

[0051] Figure 2 This is a flow chart for dynamically updating the newly added features and the correlation between various features provided in Example 1 of the present invention.

[0052] Figure 3 This is a flowchart of updating the anti-electricity theft model provided in Example 1 of the present invention.

[0053] Figure 4 A flowchart for constructing an anti-electricity theft knowledge graph provided in Example 1 of the present invention.

[0054] Figure 5 This is an anti-electricity theft knowledge graph based on the type of electricity theft caused by temporary wiring during construction, provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0056] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0057] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0058] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0059] Example 1:

[0060] like Figure 1 、 Figure 2 、 Figure 3 and Figure 4As shown, embodiment 1 of the present invention provides an anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data, including the following process:

[0061] S1: Extract new collection feature types, new collection data of various new and old collection features, and new electricity theft type sample data through the electricity consumption information collection system, and summarize the new data.

[0062] The electricity consumption information collection system collected more than 300,000 user electricity consumption samples from a certain city from July 2019 to June 2021. For relevant users, user metering indicator data was obtained from the electricity consumption information collection system, and user profile data was obtained from the marketing business application system. In addition, meteorological data was obtained from the meteorological system, combined with holiday data and various social information crawled from the Internet for collection and aggregation. The initial selection of some feature data examples is as follows:

[0063]

[0064]

[0065]

[0066] S2: Perform data preprocessing on the aggregated data, including data cleaning, data deletion, data completion and normalization.

[0067] S3: Perform multi-factor canonical correlation analysis based on the newly collected data, explore and add new electricity consumption behavior characteristics, and use the block incremental learning algorithm to update the canonical correlation analysis feature correlation coefficient value based on the newly collected data of various new and old collection characteristics.

[0068] S3.1: The sample data obtained from the electricity consumption information collection system and the marketing business application system is [X, Y], where X = {x1, x2, x3, ..., x p}, Y={y1,y2,y3,…,y p}, where X and Y represent the electricity consumption collection features whose correlation is to be determined, x and y represent the electricity consumption collection data values ​​at the collection time point, and p represents the collection time point;

[0069] S3.2: Use the canonical correlation analysis method to calculate the overall linear relationship between feature set X and feature set Y. The correlation calculation process is as follows:

[0070]

[0071] u=A T X=a1x1+a2x2+…+a p x p (2)

[0072] v=B TY=b1y1+b2y2+…+b p y p (3)

[0073] Where A={a1,a2,…,a p} and B={b1,b2,…,b p} are projection weight matrices, projections u and v are typical components, ρ represents the maximum typical correlation coefficient of u and v, C u,v is the covariance matrix of u and v, C u,u and C v,v are the autovariance matrices of u and v respectively.

[0074] In order to obtain the maximum correlation coefficient ρ and the corresponding projection weight matrices A and B, the singular value decomposition algorithm is used to solve the problem. The process is as follows:

[0075] Calculate the variance C of X and Y respectively X,X 、C Y,Y , the covariance C of X and Y X,Y , calculate the matrix M:

[0076]

[0077] Perform singular value decomposition on the matrix M to obtain the maximum singular value as the singular value vectors u and v corresponding to the first correlation coefficient ρ, then the projection weight matrices A and B are:

[0078]

[0079] Perform multi-factor canonical correlation analysis on the newly collected data to obtain its canonical correlation as the new electricity usage behavior features of the anti-electricity theft model. Examples of some of the new features obtained in this embodiment are as follows:

[0080]

[0081]

[0082] S3.3: Divide the newly collected data samples of each type of new and old collected features into incremental blocks of the same size according to the same time window. The number of samples in each block is p, as shown below:

[0083] Sample data X:

[0084] Sample data Y:

[0085] S4: Iteratively update the projection vector based on each data block. Since the projection weight matrices A and B in S3.2 are scale invariant, we define Let be a unit vector, Then ρ=||w||, and we can get By shifting terms and iterative recursion, the iteration formula for the projection vector of the qth batch is:

[0086]

[0087] in, is the initial projection weight matrix obtained in step S3.2.

[0088] The new sample data and the previous batch projection weight matrix are used to obtain the current batch projection weight matrix, thereby reducing the amount of calculation;

[0089] S3.5: Projection weight matrix based on the current batch According to formula (1), the maximum correlation coefficient ρ(q) of the newly added batch of collected data is calculated, and the maximum correlation coefficient between the newly added features and the updated features is output. Some examples are as follows:

[0090]

[0091]

[0092] S4: Based on the newly acquired features and the updated canonical correlation analysis coefficient values, the original logistic regression algorithm is used to construct a new anti-electricity theft model class based on the newly added electricity theft types, and the original anti-electricity theft model parameters are updated and optimized.

[0093] S4.1: Based on the newly added electricity consumption behavior features of the anti-electricity theft model obtained in S3 and the maximum correlation coefficient ρ(q) of each feature of the newly added batch collected data, the sample set data is added with a new vector definition Z′=(z′1,z′2,z′3,…), the new feature weight vector is θ′=(θ′1,θ′2,θ′3,…), and the objective function of the anti-electricity theft model is f(θ)=θ T *Z=(θ0,θ′) T *(Z0, Z′), category set C = {C1, C2}, C1 is the electricity theft category, initialization iteration number k = 0, allowable error ε>0, θ0 and Z0 are the original features and corresponding feature weights respectively, the feature weights are retained as the initialization weights, and the new feature weight θ′ is randomly assigned;

[0094] Among them, various anti-electricity theft models constructed based on the logistic regression algorithm are single-category anti-electricity theft models with different input features and different parameter values ​​constructed according to different electricity theft categories;

[0095] S4.2: Based on the newly added feature data and updated correlation coefficients, an iterative algorithm is used to calculate and update the model parameters:

[0096] Calculate the Hessian matrix Hk+1 :

[0097]

[0098] Among them, p (k) =θ (k+1) -θ (k) , Let the Hessian matrix H1 be the identity matrix.

[0099] The gradient descent direction is:

[0100]

[0101] Then from z (k) Departure, direction d (k) Find the step size λ k , the solution equation is:

[0102]

[0103] Update the feature weight vector θ = (θ0, θ′):

[0104] θ (k+1) =θ (k) +λ k d (k) (10);

[0105] S4.3: Substitute the updated model parameters θ into the objective function f(θ) to determine whether the error between the objective function and the actual electricity theft identification is less than the error threshold:

[0106]

[0107] If yes, go to step S4.4, otherwise return to step S4.3;

[0108] S4.4: Fix the objective function parameters and output the various anti-electricity theft models updated this time:

[0109]

[0110] Based on the updated anti-electricity theft models, 106 historical electricity theft samples and 312 normal electricity usage samples from January to June 2021 were used to identify electricity theft. The comparison of the accuracy and precision before and after the model update is shown in the following table:

[0111]

[0112]

[0113] As can be seen from the table above, the anti-electricity theft model based on incremental data updates has improved both its accuracy and precision in identifying electricity theft.

[0114] S5: Based on the newly added and updated anti-electricity theft models of various types of electricity theft, the bottom subgraph and top subgraph are constructed, and the anti-electricity theft knowledge in the anti-electricity theft knowledge graph is processed by subgraph matching.

[0115] S5.1: Define the electricity theft category in the history record as C1={c1,c2,…,c N}, the anti-electricity theft knowledge graph is G = (V, E), where V and E represent the node set and edge set of the graph respectively. The anti-electricity theft knowledge graph is divided into the bottom subgraph and the top subgraph;

[0116] S5.2: In the process of constructing the underlying subgraph, entities and attributes are extracted based on the historically accumulated anti-electricity theft record information, and triples and multitudes are defined within the knowledge graph. To achieve the directionality of knowledge association, the internal triple of the knowledge graph is represented as (h, r, a), where h and a represent the head entity and tail attribute respectively, and r represents the relationship between entities. The multitude associated with multiple entities and virtual tail attributes is represented as (h1, h2…, r, a v ), where a set of tuples contains multiple entities h1, h2…, and a set of tuples contains only one relation r and one virtual tail attribute a v ;

[0117] S5.3: Based on the expert experience and the multiple factors obtained by S3 and the corresponding maximum correlation coefficient, construct the anti-electricity theft triple (h, r, a), where h is the entity value, r is the associated trigger constraint, its threshold is the single factor electricity consumption anomaly obtained by expert experience, and a is the suspicious attribute; based on the multiple factors obtained by S3 and the corresponding maximum correlation coefficient, construct the multi-tuple (h1, h2…, r, a v ), multiple factors are multiple entities h1,h2…,a v The value of is a suspicious attribute, r represents the associated trigger constraint condition, and its threshold is the maximum correlation coefficient value. If it is higher than the threshold, the suspicious attribute a is triggered. v , the output suspicious attributes a and a of the underlying subgraph v is the input of the top-level subgraph;

[0118] S5.4: The top-level subgraph is constructed using a multi-tuple (a1, a2…, a v1 ,a v2 …,θ1,θ2,…c), and construct an anti-electricity theft model based on different electricity theft categories in S4, with suspicious attributes a and a v The anti-electricity theft model input is used, and the electricity theft category c is used as the model output. Finally, different anti-electricity theft models are obtained for different electricity theft categories. The characteristic parameter value θ of each anti-electricity theft model is extracted as the associated attribute value. The suspicious attributes a and a are used as the vAs the head entity of the top-level subgraph, the electricity theft category is used as the tail entity of the top-level subgraph, thereby realizing the knowledge connection of the anti-electricity theft knowledge graph. The output of the top-level subgraph is the electricity theft category;

[0119] S5.5: Based on the newly added features and updated maximum correlation coefficient values ​​obtained in S3, the newly added features are added to each anti-electricity theft model as feature input, and the multi-factor associated virtual attribute feature values ​​are updated based on the updated maximum correlation coefficient values. According to S4, each anti-electricity theft model is updated, and then the feature parameters of each model are updated. Based on the model parameter values, the features of each model are added and eliminated, including the addition or removal of new features and the retention or deletion of original features. The associated information of the knowledge graph is updated according to the updated feature parameters of each model, such as Figure 5 The figure below shows a knowledge graph constructed based on the types of electricity theft caused by temporary wiring during construction. The top subgraph consists of rectangles and ellipse entities, while the bottom subgraph consists of ellipse and circle entities. This helps identify the hourly time range of temporary electricity theft, thereby assisting on-site inspections.

[0120] S5.6: Based on the continuously optimized knowledge graph, anti-electricity theft behavior identification and electricity theft behavior traceability analysis are carried out.

[0121] S6: After obtaining more electricity theft-related knowledge about suspected electricity theft users through knowledge graph learning, the knowledge is organized through natural language to output an electricity theft analysis report, which increases the interpretability of the electricity theft model and serves as an auxiliary reference for power workers when conducting on-site inspections.

[0122] Example 2:

[0123] Embodiment 2 of the present invention provides an anti-electricity theft knowledge graph optimization system based on dynamic update of incremental data, including:

[0124] The data acquisition module is configured to: acquire newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types;

[0125] The preprocessing module is configured to: preprocess the acquired data;

[0126] The correlation calculation module is configured to: perform a multi-factor canonical correlation analysis based on the newly collected data to obtain new electricity consumption behavior characteristics, and use a block incremental learning algorithm to update the characteristic correlation coefficient value of the canonical correlation analysis;

[0127] The model building and optimization module is configured to: based on the newly acquired electricity consumption behavior characteristics and the updated canonical correlation analysis correlation coefficient value, use a logistic regression algorithm to build a new anti-electricity theft model class based on the newly added electricity theft type, and update and optimize the parameters of the original anti-electricity theft model;

[0128] The knowledge graph optimization module is configured to: construct bottom-level subgraphs and top-level subgraphs based on various new and updated anti-electricity theft models of electricity theft types, use subgraph matching to fuse and process the anti-electricity theft knowledge in the anti-electricity theft knowledge graph, and obtain an optimized anti-electricity theft knowledge graph.

[0129] The working method of the system is the same as the anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data provided in Example 1, and will not be repeated here.

[0130] Example 3:

[0131] Embodiment 3 of the present invention provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the steps of the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in Embodiment 1 of the present invention are implemented.

[0132] Example 4:

[0133] Embodiment 4 of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in Embodiment 1 of the present invention are implemented.

[0134] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0135] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0136] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0138] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0139] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data, The process includes: Obtain newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types; Preprocess the acquired data; Based on the newly collected data, multi-factor canonical correlation analysis is performed to obtain the new electricity consumption behavior characteristics, and the block incremental learning algorithm is used to update the characteristic correlation coefficient value of the canonical correlation analysis; Based on the newly acquired electricity consumption behavior characteristics and the updated canonical correlation analysis correlation coefficient values, a logistic regression algorithm is used to build a new anti-electricity theft model class based on the newly added electricity theft types, and the parameters of the original anti-electricity theft model are updated and optimized; Based on the newly added and updated anti-electricity theft models of various types of electricity theft, the bottom subgraph and top subgraph are constructed, and the anti-electricity theft knowledge in the anti-electricity theft knowledge graph is fused and processed by subgraph matching to obtain the optimized anti-electricity theft knowledge graph, including: Define the electricity theft category and anti-electricity theft knowledge graph in historical records, and divide the anti-electricity theft knowledge graph into the bottom subgraph and the top subgraph; In the process of constructing the underlying subgraph, entities and attributes are extracted based on the historically accumulated anti-electricity theft record information, and triples and multi-tuples are defined within the knowledge graph; Based on expert experience, the acquired multiple factors and the corresponding maximum correlation coefficient, anti-electricity theft triples and multi-tuples are constructed to complete the associated trigger constraints and suspicious attribute output as the input of the top-level subgraph; The top-level subgraph is constructed using multi-tuples. Based on the constructed anti-electricity theft model based on different electricity theft categories, suspicious attributes are used as the input of the anti-electricity theft model, and the electricity theft category is used as the model output. Different anti-electricity theft models are obtained for different electricity theft categories. The characteristic parameter values ​​of each anti-electricity theft model are extracted as the associated attribute values. The suspicious attributes are used as the head entity of the top-level subgraph, and the electricity theft category is used as the tail entity of the top-level subgraph. The output of the top-level subgraph is the electricity theft category. Based on the acquired new features and the updated maximum correlation coefficient value, the new features are added to each anti-electricity theft model as feature input, and the multi-factor associated virtual attribute feature value is updated based on the updated maximum correlation coefficient value, and each anti-electricity theft model is updated, and then the feature parameters of each model are updated, and based on the model parameter value, the features of each model are added and eliminated, and the associated information of the knowledge graph is updated according to the updated feature parameters of each model.

2. The anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data according to claim 1 comprises the following process: The preprocessing includes: data cleaning, data deletion, data completion and normalization.

3. The anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data as claimed in claim 1, The process includes: Based on the newly collected data, a multi-factor canonical correlation analysis is performed to obtain the newly added electricity consumption behavior characteristics, and the block incremental learning algorithm is used to update the characteristic correlation coefficient value of the canonical correlation analysis, including: Based on the newly acquired data of the newly acquired features, the multi-correlation factor canonical correlation analysis method is used to calculate the overall linear relationship between the sample feature sets, and the canonical correlation is obtained as the newly acquired electricity consumption behavior feature of the anti-electricity theft model; Divide the collected data of various new and old collected features into incremental data blocks of the same size according to the same time window; Using the projection weight matrix of the incremental data block and the previous batch of incremental data blocks, the projection vector is iteratively updated to obtain the projection weight matrix of the incremental data block newly added in the current batch; Based on the projection weight matrix of the incremental data blocks newly added in the current batch, the maximum correlation coefficient of the newly added batch of collected data is obtained, and the maximum correlation coefficient between the newly added features and the updated features is obtained.

4. The anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data as claimed in claim 3, The process includes: Based on the newly acquired electricity consumption behavior characteristics and the updated typical correlation analysis correlation coefficient value, a logistic regression algorithm is used to build a new anti-electricity theft model class based on the newly added electricity theft type, and the parameters of the original anti-electricity theft model are updated and optimized, including: Based on the newly acquired electricity consumption behavior characteristics and the maximum correlation coefficient of each characteristic of the newly acquired batch collection data, data definition is performed; Initialize the anti-electricity theft model parameters of the newly added electricity theft categories based on the newly added electricity theft types, and initialize the parameters of various anti-electricity theft models built based on the logistic regression algorithm; Based on the newly added feature data and the updated maximum correlation coefficient, an iterative algorithm is used to calculate and update each model parameter; Substitute the updated model parameters into the objective function to determine whether the error between the objective function and the actual electricity theft identification is less than the error threshold. If yes, proceed to the next step, otherwise return to the previous step. The objective function parameters are fixed, and various anti-electricity theft models updated this time are output.

5. The anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data according to claim 4 comprises the following process: Various types of anti-electricity theft models constructed based on the logistic regression algorithm are single-category anti-electricity theft models with different input features and different parameter values ​​constructed according to different electricity theft categories.

6. The anti-electricity theft knowledge graph optimization method based on dynamic update of incremental data according to claim 1 comprises the following process: Add and eliminate features of each model, including: Add or remove new features, and retain or delete existing features. 7.An anti-electricity theft knowledge graph optimization system based on dynamic update of incremental data, comprising: The data acquisition module is configured to: acquire newly added collection feature types, newly added collection data of various new and old collection features, and newly added electricity theft types; The preprocessing module is configured to: preprocess the acquired data; The correlation calculation module is configured to: perform a multi-factor typical correlation analysis based on the newly collected data to obtain new electricity consumption behavior characteristics, and use a block incremental learning algorithm to update the characteristic correlation coefficient value of the typical correlation analysis; The model building and optimization module is configured to: based on the acquired new electricity consumption behavior characteristics and the updated typical correlation analysis correlation coefficient value, use the logistic regression algorithm to build a new anti-electricity theft model class based on the newly added electricity theft type, and update and optimize the parameters of the original anti-electricity theft model; The knowledge graph optimization module is configured to: construct bottom-level subgraphs and top-level subgraphs based on various new and updated anti-electricity theft models of electricity theft types, fuse and process the anti-electricity theft knowledge in the anti-electricity theft knowledge graph using subgraph matching, and obtain an optimized anti-electricity theft knowledge graph; The knowledge graph optimization module specifically includes: Define the electricity theft category and anti-electricity theft knowledge graph in historical records, and divide the anti-electricity theft knowledge graph into the bottom subgraph and the top subgraph; In the process of constructing the underlying subgraph, entities and attributes are extracted based on the historically accumulated anti-electricity theft record information, and triples and multi-tuples are defined within the knowledge graph; Based on expert experience, the acquired multiple factors and the corresponding maximum correlation coefficient, anti-electricity theft triples and multi-tuples are constructed to complete the associated trigger constraints and suspicious attribute output as the input of the top-level subgraph; The top-level subgraph is constructed using multi-tuples. Based on the constructed anti-electricity theft model based on different electricity theft categories, suspicious attributes are used as the input of the anti-electricity theft model, and the electricity theft category is used as the model output. Different anti-electricity theft models are obtained for different electricity theft categories. The characteristic parameter values ​​of each anti-electricity theft model are extracted as the associated attribute values. The suspicious attributes are used as the head entity of the top-level subgraph, and the electricity theft category is used as the tail entity of the top-level subgraph. The output of the top-level subgraph is the electricity theft category. Based on the acquired new features and the updated maximum correlation coefficient value, the new features are added to each anti-electricity theft model as feature input, and the multi-factor associated virtual attribute feature value is updated based on the updated maximum correlation coefficient value, and each anti-electricity theft model is updated, and then the feature parameters of each model are updated, and based on the model parameter value, the features of each model are added and eliminated, and the associated information of the knowledge graph is updated according to the updated feature parameters of each model.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the anti-electricity theft knowledge graph optimization method based on dynamic updating of incremental data as described in any one of claims 1 to 6 are implemented.

Citation Information

Patent Citations

  • Anti-electricity-stealing analysis warning method for analyzing and optimizing model based on logistic regression probability

    CN107145966A

  • Electricity larceny prevention simulation monitoring method, system and equipment for re-electrified energy consumption device, and medium

    CN110083986A

  • Electricity larceny prevention analysis method based on typical industry feature library and electricity larceny prevention sample library

    CN110223196A

  • Multi-dimensional intelligent analysis experience algorithm and classification algorithm for judging default electricity utilization and electricity stealing

    CN113392910A

  • Predictive theft notification for the prevention of theft

    US20140254890A1