An electricity bill anomaly detection method, device, and equipment of a communication base station and a medium

Abstract

The application relates to an electricity fee anomaly detection method, device and equipment of a communication base station and a medium. The application obtains working parameters, electricity fee rate data and base station working service logic data of a plurality of communication base stations in a target detection area; constructs a base station correlation graph according to the base station working service logic data; performs weighted aggregation on the working parameters, the electricity fee rate data and the base station correlation graph through a graph attention network to obtain aggregation features, and determines latent variables according to the aggregation features; inputs the latent variables into a feature reconstruction decoder and an electricity fee rate prediction decoder respectively to obtain feature reconstruction results and electricity fee rate prediction results; obtains electricity fee anomaly scores according to the feature reconstruction results and the electricity fee rate prediction results, and determines electricity fee anomaly base stations and electricity fee anomaly types according to the electricity fee anomaly scores. The scheme in the application can improve the sensitivity to electricity fee anomalies such as rate configuration errors, thereby reducing false positives and improving detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of communication network operation and maintenance and data intelligence technology, specifically to a method, device, equipment and medium for detecting abnormal electricity charges of communication base stations. Background Technology

[0002] As the core infrastructure of mobile communication networks, communication base stations play a crucial role in wireless signal coverage, data transmission, and service delivery. Their electricity costs are a significant component of the operating costs of telecommunications operators. During the operation of communication base stations, various factors such as equipment malfunctions, metering device abnormalities, deviations in electricity rate configurations, abnormal line losses, and illegal electricity use can cause base station electricity cost data to deviate from the normal range, directly affecting the accuracy of electricity cost accounting and the effectiveness of energy consumption management. Therefore, it is necessary to detect abnormal electricity costs of communication base stations to ensure compliance in electricity cost auditing and efficient energy consumption management.

[0003] Currently, conventional methods for detecting abnormal electricity charges at communication base stations mainly employ threshold rule-based judgment, year-on-year and month-on-month statistical comparisons of electricity charges and energy consumption data, or constructing shallow regression and classification models for detection by treating individual base stations as independent samples. These conventional methods only analyze data from a single base station and cannot cross-validate with the energy consumption and load trends of related base stations. This easily leads to misjudging normal service fluctuations and seasonal load changes as abnormal electricity charges. Furthermore, electricity charge multipliers are affected by factors such as equipment configuration, metering parameters, and ledger registration, exhibiting non-linear, multimodal, and heterogeneous characteristics. Shallow models based solely on data from a single base station struggle to capture the complex relationship between energy consumption and multipliers, are insensitive to core anomalies such as multiplier configuration errors, and are prone to missed detection. Moreover, the accuracy of anomaly detection is insufficient, requiring extensive manual review, ultimately resulting in a high false alarm rate and low efficiency in electricity charge auditing.

[0004] Therefore, improving the sensitivity to electricity bill anomalies such as incorrect multiplier configuration, thereby reducing false alarms and improving detection efficiency, is an urgent problem to be solved. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to improve the sensitivity to electricity bill anomalies such as incorrect multiplier configuration, thereby reducing false alarms and improving detection efficiency.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: This application provides a method for detecting abnormal electricity charges in a communication base station, including: Acquire the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area; Based on the base station operational service logic data, a base station association diagram is constructed; The working parameters, the electricity rate data, and the base station association graph are weighted and aggregated using a graph attention network to obtain aggregated features, and latent variables are determined based on the aggregated features. The latent variables are input into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result; Based on the feature reconstruction results and the electricity cost ratio prediction results, an electricity cost anomaly score is obtained, and based on the electricity cost anomaly score, the base station with electricity cost anomaly and the type of electricity cost anomaly are determined.

[0007] The beneficial effects of this invention are as follows: A base station association graph is constructed based on the base station's operational service logic data. This graph breaks the limitation of treating base stations as independent entities, utilizing relationships between base stations to uncover potential anomalies and improve sensitivity to abnormal situations. By using a graph attention network to weight and aggregate various types of data and determine latent variables, potential connections and features between data can be uncovered. These latent variables are input into a feature reconstruction decoder and an electricity rate prediction decoder, respectively, to obtain feature reconstruction results and electricity rate prediction results, which are then analyzed from two dimensions. Based on these two results, an electricity rate anomaly score is obtained, and the base station and type of electricity rate anomaly are determined accordingly. Through multi-dimensional analysis and comprehensive judgment, electricity rate configuration errors and other electricity rate anomalies can be identified more comprehensively and accurately, thereby improving sensitivity to these anomalies, reducing false alarms, and increasing detection efficiency.

[0008] Based on the above technical solution, the present invention can be further improved as follows.

[0009] Furthermore, the base station operational service logic data includes station affiliation, power supply affiliation, and geographical proximity. The step of constructing a base station association graph based on the base station operational service logic data includes: Each of the aforementioned communication base stations is treated as a node; Based on the station affiliation relationship, an undirected edge is established for the nodes corresponding to two communication base stations with the same station identifier; Based on the power supply attribution relationship, an undirected edge is established for nodes corresponding to two communication base stations that are at the same power supply source; Based on the geographical proximity relationship, an undirected edge is established for nodes corresponding to two communication base stations whose geographical distance is less than a preset distance threshold; The base station association graph is determined based on the nodes and the undirected edges.

[0010] The beneficial effects of adopting the above-mentioned further solutions are as follows: The station affiliation relationship connects base stations belonging to the same station in the graph, helping to uncover potential collaborative working patterns or abnormal associations between base stations within the same station, and improving sensitivity to anomalies related to the same station; the power supply affiliation relationship links base stations with the same power source, enabling effective detection of power supply-related anomalies such as power line faults and abnormal power distribution; the geographical proximity relationship connects geographically close base stations, considering that geographically close locations may face similar environmental factors affecting base station energy consumption, which can help discover anomalies caused by environmental factors. The base station association graph constructed by integrating these relationships more comprehensively reflects the potential connections between base stations, thereby improving the accuracy and efficiency of electricity fee anomaly detection, reducing false alarm rates, and more accurately identifying various electricity fee anomalies related to base station business logic.

[0011] Furthermore, the step of using a graph attention network to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph to obtain aggregated features, and determining latent variables based on the aggregated features, includes: Based on the operating parameters and the electricity rate data, construct a node feature matrix; Based on the base station association graph, determine the node adjacency matrix; The node feature matrix and the node adjacency matrix are transmitted to the graph attention network to perform weighted aggregation of the contributions of the multiple communication base stations through attention coefficients to obtain the aggregated features, wherein the attention coefficients are obtained by nonlinear transformation and normalization of the node feature matrix; The aggregated features are input into a variational autoencoder to encode the latent distribution parameters. The latent distribution parameters are reparameterized to obtain the latent variables.

[0012] The beneficial effects of adopting the above-mentioned further scheme are as follows: By constructing a node feature matrix and a node adjacency matrix, the data features and connection relationships of base stations are integrated respectively, providing a comprehensive and orderly input for the graph attention network. Using the attention coefficients obtained by nonlinear transformation and normalization of the node feature matrix, the contributions of multiple communication base stations are weighted and aggregated, enabling the model to dynamically focus on the importance of different base stations in the overall picture, effectively mining the complex data relationships and potential features between base stations, and obtaining more representative aggregated features. The aggregated features are input into a variational autoencoder and reparameterized to determine latent variables. The variational autoencoder can learn the latent distribution of the data, and reparameterization solves the differentiation problem in the sampling process, facilitating model training and optimization. The final latent variables contain the key information of the original data, providing high-quality basic data for feature reconstruction and decoding prediction. Therefore, in electricity fee anomaly detection, it is possible to more accurately capture abnormal patterns in the data, improving the accuracy and reliability of anomaly detection.

[0013] Furthermore, the feature reconstruction result includes reconstructed features, the electricity bill multiplier prediction result includes an electricity bill prediction multiplier, and the step of obtaining an electricity bill anomaly score based on the feature reconstruction result and the electricity bill multiplier prediction result includes: The reconstructed features are compared with the original features corresponding to the working parameters to obtain the feature reconstruction error; The predicted electricity price multiplier is compared with the actual electricity price multiplier data to obtain the electricity price multiplier deviation. The electricity bill anomaly score is obtained based on the feature reconstruction error and the electricity bill multiplier deviation.

[0014] The beneficial effects of adopting the above-mentioned further scheme are as follows: By separately obtaining the feature reconstruction error by comparing the reconstructed features with the original features, and the electricity charge multiplier deviation by comparing the predicted electricity charge multiplier with the actual electricity charge multiplier data, an electricity charge anomaly score is obtained based on these two deviations. The feature reconstruction error reflects the accuracy of the model's learning of the base station's operating parameters. If the error is large, it may mean that the base station's operating status is abnormal or there are problems in the data acquisition and processing process. The electricity charge multiplier deviation directly reflects the difference between the predicted electricity charge multiplier and the actual multiplier, and can effectively identify anomalies such as incorrect electricity charge multiplier configuration. By comprehensively calculating the electricity charge anomaly score from these two deviations, and considering the two key dimensions of operating parameters and electricity charge multiplier, the measurement of base station electricity charge anomalies becomes more comprehensive and accurate, enabling more precise location of abnormal base stations and effectively improving the detection sensitivity of electricity charge anomalies.

[0015] Furthermore, obtaining the electricity bill anomaly score based on the feature reconstruction error and the electricity bill multiplier deviation includes: Based on the historical false alarm rate or the cost of detecting abnormal electricity charges, determine the weighting coefficients corresponding to the feature reconstruction error and the deviation of the electricity charge multiplier. Based on the weighting coefficients, the feature reconstruction error and the electricity rate deviation are weighted and summed to obtain the basic anomaly score; Based on the potential distribution parameters, determine the potential distribution distance term or confidence term; The basic anomaly score is fused with the potential distribution distance term or the confidence term to obtain the electricity bill anomaly score.

[0016] The beneficial effects of adopting the above-mentioned further scheme are as follows: Determining the weighting coefficients based on historical false alarm rates or electricity bill anomaly detection costs allows for flexible adjustment of the importance of feature reconstruction errors and electricity bill ratio deviations in anomaly score calculation, making the anomaly score more aligned with actual needs and reducing false alarms or missed alarms caused by fixed weights. The weighted summation of the basic anomaly score initially integrates the impact of two key deviations on anomalies, considering both base station operating parameters and electricity bill ratios, thus improving the accuracy of anomaly detection. Furthermore, determining the potential distribution distance or confidence term based on potential distribution parameters and integrating it with the basic anomaly score further introduces information at the potential distribution level, providing an additional reference dimension for anomaly judgment. This ensures that the final electricity bill anomaly score not only considers direct deviations but also incorporates the relationship between the potential distribution and the normal distribution, as well as the credibility of the anomaly judgment, thereby more comprehensively and accurately reflecting the degree of electricity bill anomalies at the base station, significantly improving the sensitivity and accuracy of electricity bill anomaly detection.

[0017] Furthermore, determining the abnormal electricity bill base station and the type of abnormal electricity bill based on the abnormal electricity bill score includes: Obtain the preset threshold for abnormal electricity bill scores; The abnormal electricity charges score of each communication base station is compared with the abnormal electricity charges score threshold, and communication base stations with an abnormal electricity charges score greater than the abnormal electricity charges score threshold are selected as the abnormal electricity charges base stations. Calculate the proportion of the feature reconstruction error and the electricity rate deviation corresponding to the abnormal electricity charges base station in the abnormal electricity charge score; The type of electricity bill anomaly is determined based on the stated percentage.

[0018] The beneficial effects of adopting the above-mentioned further scheme are as follows: By setting a preset threshold for electricity fee anomaly scores and comparing the electricity fee anomaly scores of each base station with this threshold, base stations that may have electricity fee anomalies can be quickly and effectively screened out, narrowing the scope of anomaly investigation and improving detection efficiency. Further calculation of the proportion of feature reconstruction error and electricity fee ratio deviation in the electricity fee anomaly score, and determination of the electricity fee anomaly type based on this proportion, makes the analysis of anomalies more in-depth and detailed. This method can not only determine whether a base station has electricity fee anomalies, but also preliminarily locate the root cause of the anomaly, whether it is related to the base station's operating parameters or the electricity fee ratio, providing a clear direction for targeted solutions to electricity fee anomaly problems. This improves the accuracy and practicality of the entire electricity fee anomaly detection system, effectively avoids blind investigation, and saves human and material resources.

[0019] Furthermore, obtaining the preset threshold for abnormal electricity charges includes: Obtain historical base station electricity fee detection data and manual detection work order data to obtain a historical sample dataset; Based on the abnormal score distribution in the historical sample dataset, determine the initial score threshold corresponding to the specified quantile; The initial score threshold is adjusted based on the recall rate and detection accuracy constraints of the electricity bill anomaly detection. The adjusted threshold will be used as the preset threshold for abnormal electricity charges.

[0020] The beneficial effects of adopting the above-mentioned further solution are as follows: By acquiring historical base station electricity fee detection data and manual detection work order data, a historical sample dataset is constructed. The distribution of abnormal scores within this dataset is used to determine the initial score threshold. Based on historical experience, a preliminary anomaly judgment standard with a certain reference value is set, which can reflect the anomaly distribution characteristics in past detection situations. Then, the initial score threshold is adjusted according to the recall rate and detection accuracy constraints of electricity fee anomaly detection, so that the threshold can adapt to actual business needs and balance the false positives and false negatives in the detection process. On the one hand, by reasonably adjusting the threshold to improve the recall rate, more base stations with actual electricity fee anomalies can be detected, avoiding false negatives; on the other hand, considering detection accuracy can effectively reduce false positives, making the detection results more reliable. The finally determined preset electricity fee anomaly score threshold, which integrates historical data and business requirements, can more accurately screen out truly abnormal electricity fee base stations, thereby improving the accuracy and reliability of the entire electricity fee anomaly detection system, optimizing the anomaly detection process, and improving detection efficiency and quality.

[0021] This application embodiment also provides a device for detecting abnormal electricity charges in a communication base station, the device comprising: The acquisition module is used to acquire the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area. The construction module is used to construct a base station association diagram based on the base station working service logic data; The aggregation module is used to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph through a graph attention network to obtain aggregated features, and to determine potential variables based on the aggregated features; The decoding module is used to input the latent variables into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result; The detection module is used to obtain an electricity charge anomaly score based on the feature reconstruction result and the electricity charge multiplier prediction result, and to determine the electricity charge anomaly base station and the electricity charge anomaly type based on the electricity charge anomaly score.

[0022] This application also provides an electronic device, including a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute the steps in any of the electricity bill anomaly detection methods for communication base stations provided in this application.

[0023] This application also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute steps in any of the electricity fee anomaly detection methods for communication base stations provided in this application. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the method for detecting abnormal electricity charges in a communication base station according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the communication base station electricity fee anomaly detection device provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0025] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0026] This application provides a method, apparatus, electronic device, and medium for detecting abnormal electricity charges in a communication base station.

[0027] Specifically, the electricity fee anomaly detection device of the communication base station can be integrated into an electronic device, such as a terminal or server. The terminal can be a mobile phone, tablet, smart Bluetooth device, laptop, or personal computer (PC); the server can be a single server or a server cluster composed of multiple servers.

[0028] In some embodiments, the electricity fee anomaly detection device for the communication base station can also be integrated into multiple electronic devices. For example, the electricity fee anomaly detection device for the communication base station can be integrated into multiple servers, and the electricity fee anomaly detection method for the communication base station of this application can be implemented by multiple servers.

[0029] In some embodiments, the terminal can also be used as a server to implement some or all of the functions of a server.

[0030] In the embodiments of this application, the terms "module" or "unit" can refer to a computer program or part of a computer program with a predetermined function, which works together with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0031] The following sections provide detailed descriptions of each example. It should be noted that the order of the following embodiments is not intended to limit the preferred order of the embodiments.

[0032] This embodiment provides a method for detecting abnormal electricity charges in a communication base station. See also... Figure 1 , Figure 1 This is a flowchart illustrating the method for detecting abnormal electricity charges in a communication base station according to an embodiment of the present invention. Figure 1 As shown, the specific process of the electricity fee anomaly detection method for this communication base station can be as follows: 110. Obtain the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area.

[0033] Among them, a communication base station is a type of radio station, which refers to a radio transceiver station that transmits information between a mobile communication switching center and a mobile phone terminal within a certain radio coverage area.

[0034] Among them, the operating parameters include various indicator data during the operation of the communication base station, such as power, signal strength, communication traffic, and equipment runtime. These parameters reflect the real-time operating status of the base station and are crucial for analyzing whether the base station is operating normally and its energy consumption.

[0035] Among them, the electricity charge multiplier data is used to calculate the electricity cost of the base station. It reflects the conversion ratio between the actual electricity consumption and the billed electricity consumption, and is directly related to the calculation of the base station's electricity cost.

[0036] Among them, base station working business logic data describes the business-level relationship between base stations. For example, the local station affiliation relationship indicates that the base station belongs to the same local station; the power supply affiliation relationship reflects that the base stations have the same power supply head; the geographical proximity relationship indicates the degree of proximity of the base stations in terms of geographical location, etc. Through these relationships, potential connections between base stations can be explored.

[0037] For example, the specific methods for obtaining operating parameters, electricity rate data, and base station operating business logic data can be as follows: collect operating parameters, electricity rate data, and base station operating business logic data of multiple base stations within the target detection area through data acquisition systems and databases related to communication base stations.

[0038] 120. Construct a base station association diagram based on the base station working service logic data.

[0039] Among them, the base station association graph is a graph structure composed of communication base stations as nodes and the connection relationship (edge) between nodes determined according to the base station working service logic data. It intuitively shows the service association between base stations.

[0040] For example, the base station association graph can be constructed in the following ways: Each communication base station is treated as a node. Based on business logic such as local station affiliation, power supply affiliation, and geographical proximity, connections are established between nodes to form a base station association diagram, which intuitively displays the business connections between base stations.

[0041] 130. Using a graph attention network, the operating parameters, the electricity rate data, and the base station association graph are weighted and aggregated to obtain aggregated features, and latent variables are determined based on the aggregated features.

[0042] Among them, graph attention network is a neural network designed for graph structure data. It can dynamically weight the importance of different neighboring nodes according to node characteristics through attention mechanism, thereby mining potential features and relationships in graph data.

[0043] Among them, the aggregated feature is a feature representation obtained by weighting and aggregating the working parameters, electricity rate data and base station association graph through a graph attention network. It integrates key information from multiple data sources and contains potential features related to abnormal base station electricity rates.

[0044] Among them, the latent variable is obtained by encoding the aggregated features through a variational autoencoder and reparameterizing. It is a latent representation of the original data features, containing key information of the data, and is used for subsequent decoding operations.

[0045] 140. Input the latent variables into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result.

[0046] Among them, the feature reconstruction decoder is a neural network structure that takes latent variables as input and attempts to reconstruct features similar to the original working parameters. By comparing the reconstructed features with the original features, the model's ability to learn and reconstruct the original data can be judged, and any anomalies in the working parameters can be detected.

[0047] Among them, the electricity rate prediction decoder takes the latent variable as input, predicts the electricity rate value, and compares it with the actual electricity rate data to determine whether there is an anomaly in the electricity rate.

[0048] Among them, the feature reconstruction result is the reconstructed feature output by the feature reconstruction decoder based on the latent variables, which reflects the model's restoration of the base station's operating parameters.

[0049] Among them, the electricity rate prediction result is the predicted electricity rate output by the electricity rate prediction decoder based on the latent variables, which is used for comparative analysis with the actual electricity rate.

[0050] 150. Based on the feature reconstruction results and the electricity fee multiplier prediction results, obtain the electricity fee anomaly score, and determine the electricity fee anomaly base station and the electricity fee anomaly type based on the electricity fee anomaly score.

[0051] Among them, the electricity charge anomaly score is a quantitative value obtained by combining the comprehensive feature reconstruction results and the electricity charge multiplier prediction results. It is used to measure the degree of electricity charge anomalies of communication base stations. The higher the score, the greater the possibility of anomalies.

[0052] Among them, the base stations with abnormal electricity charges are identified based on their abnormal electricity charge scores, and these are communication base stations whose electricity charges may be abnormal.

[0053] Among them, the electricity bill anomaly type is a specific category of communication base station electricity bill anomalies determined based on factors such as the proportion of feature reconstruction error and electricity bill multiplier deviation in the electricity bill anomaly score, such as electricity bill anomalies caused by abnormal operating parameters, anomalies caused by incorrect electricity bill multiplier configuration, etc.

[0054] For example, the target detection area is a specific urban area of ​​a city, containing 5 communication base stations. Base station 1's operating parameters include a power of 30 kW, a signal strength of 90 dBm, and a monthly communication traffic of 500 GB, with an electricity cost multiplier of 1.15. Base station operational logic data shows that base station 1 and base station 2 belong to the same local exchange, base station 3 and base station 4 are powered by the same transformer, and base station 5 is geographically less than 200 meters away from base station 1. First, the operating parameters, electricity cost multiplier data, and operational logic data of these base stations are obtained. Then, a base station association graph is constructed based on the operational logic data, treating each base station as a node. Edges are connected between corresponding nodes of base station 1 and base station 2, between corresponding nodes of base station 3 and base station 4, and between corresponding nodes of base station 5 and base station 1. Next, the operating parameters and electricity cost multiplier data are used to construct a node feature matrix, and the base station association graph is transformed into a node adjacency matrix. This matrix is ​​then input into a graph attention network for weighted aggregation to obtain aggregated features, and latent variables are further determined. The latent variables are input into the feature reconstruction decoder and the electricity rate prediction decoder, respectively. The power in the reconstructed features output by the feature reconstruction decoder is 28 kW, which is compared with the original power of 30 kW to obtain the feature reconstruction error. The electricity rate prediction decoder predicts an electricity rate of 1.2, which is compared with the actual electricity rate of 1.15 to obtain the electricity rate deviation. An electricity rate anomaly score is calculated using a preset weight of 0.6 for the feature reconstruction error and 0.4 for the electricity rate deviation. The preset threshold for the electricity rate anomaly score is 80. If a base station's score is greater than 80, it is identified as a base station with an electricity rate anomaly. If the feature reconstruction error accounts for a large proportion of the base station's score, it can be preliminarily determined that the anomaly type is related to operating parameters, such as abnormal power acquisition.

[0055] As can be seen from the above, in this embodiment of the application, a base station association graph is constructed based on the base station working service logic data. The base station association graph can break the limitation of treating base stations as independent individuals, utilize the relationships between base stations to mine potential anomalies, and improve the sensitivity to anomalies. By using a graph attention network to weight and aggregate various types of data and determine potential variables, potential connections and features between data can be mined. The potential variables are input into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction results and the electricity rate prediction results, which are analyzed from two dimensions. Based on these two results, an electricity rate anomaly score is obtained, and the base station and type of electricity rate anomaly are determined accordingly. Through multi-dimensional analysis and comprehensive judgment, electricity rate anomalies such as misconfiguration can be identified more comprehensively and accurately, thereby improving the sensitivity to these anomalies, reducing false alarms, and improving detection efficiency.

[0056] Preferably, in this embodiment, the base station operational service logic data includes local station affiliation, power supply affiliation, and geographical proximity. The step of constructing a base station association graph based on the base station operational service logic data includes: Each of the aforementioned communication base stations is treated as a node; Based on the station affiliation relationship, an undirected edge is established for the nodes corresponding to two communication base stations with the same station identifier; Based on the power supply attribution relationship, an undirected edge is established for nodes corresponding to two communication base stations that are at the same power supply source; Based on the geographical proximity relationship, an undirected edge is established for nodes corresponding to two communication base stations whose geographical distance is less than a preset distance threshold; The base station association graph is determined based on the nodes and the undirected edges.

[0057] In this context, station affiliation refers to the relationship of a communication base station to which it belongs at the administrative or operational management level. Each station typically has its own independent management system and resource allocation, and the station identifier is used to uniquely identify a station's number, name, and other information. Station affiliation clarifies which base stations fall under the same management scope.

[0058] Among them, the power supply attribution relationship reflects the connection between communication base stations in terms of power supply. That is, if two base stations are powered by the same electricity meter, power supply unit, transformer or distribution box, it indicates that they have a common source of power supply, which may affect their energy consumption characteristics and electricity billing methods.

[0059] Geographic proximity focuses on the closeness of communication base stations in terms of geographical location. The preset distance threshold is a distance standard set based on actual needs and experience to determine whether two base stations are geographically close. This helps to account for the fact that base stations in geographically close locations may be affected by similar environmental factors.

[0060] For example, the specific method for obtaining nodes can be as follows: In the process of constructing the base station association graph, each communication base station is abstracted as a node in the graph structure. This is the basic step in constructing the graph and prepares for determining the connection relationship between nodes in the future.

[0061] For example, an undirected edge can be constructed as follows: Traverse all base stations. For any two base stations with the same local exchange identifier, establish an undirected edge between their corresponding nodes. An undirected edge indicates that the two base stations belong to the same local exchange. This connection reflects the relevance of the base stations at the management level, meaning that they may have connections in terms of resource sharing, collaborative work, etc.

[0062] Iterate through all base stations, find two base stations that share the same power source (such as the same electricity meter, power supply unit, etc.), and establish an undirected edge between their corresponding nodes. This step highlights the correlation between base stations in terms of power supply, which may involve mutual influences in power distribution and energy consumption.

[0063] The geographical distance between any two base stations is calculated by measuring or acquiring their geographic location information. For two base stations with a geographical distance less than a preset distance threshold, an undirected edge is established between their corresponding nodes. This takes into account that geographically close base stations may face similar environmental factors, such as temperature and humidity, which will have a similar impact on base station energy consumption and electricity costs.

[0064] For example, the specific method for obtaining the base station association graph can be as follows: By integrating the previously identified nodes and the undirected edges established between them, a complete graph structure is formed, namely the base station association graph. This graph intuitively illustrates the relationships between base stations based on business logic such as local station affiliation, power supply affiliation, and geographical proximity, providing a foundation for subsequent data analysis on the graph.

[0065] For example, consider communication base stations A, B, C, D, and E. Base station A and base station B belong to the same "local station X". Based on the local station affiliation, an undirected edge is established between the nodes corresponding to base station A and base station B. Base station C and base station D are powered by the same transformer. Based on the power supply affiliation, an undirected edge is established between the nodes corresponding to base station C and base station D. The geographical distance between base station E and base station A is measured to be 300 meters, and the preset distance threshold is 500 meters. Based on geographical proximity, an undirected edge is established between the nodes corresponding to base station E and base station A. Through these operations, using base stations A, B, C, D, and E as nodes and connecting them with the established undirected edges, a base station association graph is determined.

[0066] In this embodiment, the local station affiliation relationship connects base stations belonging to the same local station with edges in the graph, which helps to uncover potential collaborative working patterns or abnormal associations between base stations within the same local station, and improves the sensitivity to anomalies related to the same local station. The power supply affiliation relationship connects base stations with the same power source, enabling effective detection of power supply-related anomalies, such as power line faults and abnormal power distribution. The geographical proximity relationship connects base stations that are geographically close, considering that geographically close base stations may face similar environmental factors affecting their energy consumption, which can help detect anomalies caused by environmental factors. The base station association graph constructed by combining these relationships more comprehensively reflects the potential connections between base stations, thereby improving the accuracy and efficiency of electricity fee anomaly detection, reducing the false alarm rate, and more accurately identifying various electricity fee anomalies related to base station business logic.

[0067] Preferably, in an embodiment, the step of using a graph attention network to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph to obtain aggregated features, and determining latent variables based on the aggregated features, includes: Based on the operating parameters and the electricity rate data, construct a node feature matrix; Based on the base station association graph, determine the node adjacency matrix; The node feature matrix and the node adjacency matrix are transmitted to the graph attention network to perform weighted aggregation of the contributions of the multiple communication base stations through attention coefficients to obtain the aggregated features, wherein the attention coefficients are obtained by nonlinear transformation and normalization of the node feature matrix; The aggregated features are input into a variational autoencoder to encode the latent distribution parameters. The latent distribution parameters are reparameterized to obtain the latent variables.

[0068] The node feature matrix is ​​a data structure that organizes the operating parameters and electricity rate data of each communication base station in matrix form. Each row or column of the matrix corresponds to the relevant data of a base station, thus integrating the data of all base stations into a single matrix. This facilitates unified processing in the graph attention network to uncover potential relationships between the data.

[0069] The node adjacency matrix, determined based on the base station association graph, describes the connectivity between base station nodes. In this matrix, if an edge connects two nodes, the corresponding element has a value of 1 (or represents the edge weight); otherwise, the element has a value of 0. It visually represents the topology of the base station association graph, providing information about the connections between base stations for the graph attention network.

[0070] The attention coefficients are values ​​obtained by performing nonlinear transformations and normalization on the node feature matrix. These coefficients measure the importance of different neighboring base stations in the weighted aggregation process, i.e., the contribution of each base station to the aggregated features. Nonlinear transformations enhance the model's ability to capture complex data relationships, while normalization ensures comparability between different coefficients.

[0071] The aggregated feature is a feature representation obtained by weighting and aggregating operating parameters, electricity rate data, and base station association graphs using a graph attention network. It integrates data information from multiple base stations, considers the connectivity between base stations, and includes potential features related to abnormal base station electricity rates.

[0072] The latent distribution parameters, obtained by encoding the aggregated features using a variational autoencoder, describe the distribution of latent variables and can be represented by statistical parameters such as mean and variance. These parameters reflect the distribution characteristics of the aggregated features in the latent space and are an important basis for determining latent variables.

[0073] Among them, latent variables are variables obtained by reparameterizing the latent distribution parameters. They are a latent representation of the original data (operating parameters, electricity rate data, etc.) and contain the core feature information of the data.

[0074] For example, the node feature matrix can be constructed as follows: The operating parameters and electricity rate data of each communication base station are arranged into rows or columns of a matrix according to certain rules. For example, the operating parameters (such as power, traffic, etc.) and electricity rate data of base station 1 are arranged in the first row of the matrix, the data of base station 2 are arranged in the second row, and so on, forming a node feature matrix containing all base station-related data.

[0075] For example, the node adjacency matrix can be determined as follows: Traverse all node pairs in the base station association graph. If there is a connecting edge between two nodes, set the element at the corresponding position in the node adjacency matrix to 1 (or set the corresponding value according to the weight of the edge). If there is no connecting edge, set it to 0. For example, if base station 1 is connected to base station 2, set the element in the 1st row and 2nd column and the 2nd row and 1st column of the node adjacency matrix to 1, thereby constructing a node adjacency matrix that reflects the topology of the base station association graph.

[0076] For example, the specific methods for obtaining aggregated features can be as follows: The constructed node feature matrix and node adjacency matrix are input into the graph attention network. The graph attention network first performs a nonlinear transformation and normalization on the node feature matrix to obtain attention coefficients. Then, based on these attention coefficients, the contributions of different neighboring base stations are weighted and aggregated. For example, for a given base station node, the features of its neighboring nodes (nodes connected to it) are weighted and summed according to the attention coefficients to obtain the aggregated features of that node. The aggregated features of all nodes together constitute the aggregated features of the entire network.

[0077] For example, the latent distribution parameters can be obtained in the following ways: The aggregated features obtained through the graph attention network are input into the variational autoencoder. The variational autoencoder encodes the aggregated features through a neural network structure, learns the distribution pattern of the aggregated features in the latent space, and finally outputs latent distribution parameters describing the distribution of latent variables, such as mean and variance.

[0078] For example, the methods for obtaining latent variables can be as follows: Since directly sampling the latent distribution parameters is inconvenient for derivative calculations during model training, a reparameterization technique is employed. This involves introducing a random variable (which could be normally distributed noise) related to the latent distribution parameters and transforming the sampling process into a differentiable operation through mathematical transformations, thereby obtaining the latent variable. This allows for convenient optimization of model parameters using the backpropagation algorithm during model training.

[0079] In this embodiment, by constructing a node feature matrix and a node adjacency matrix, the data features and connection relationships of base stations are integrated, providing a comprehensive and ordered input for the graph attention network. Attention coefficients obtained by nonlinear transformation and normalization of the node feature matrix are used to weighted aggregate the contributions of multiple communication base stations. This allows the model to dynamically focus on the importance of different base stations within the overall picture, effectively mining complex data relationships and latent features between base stations, and obtaining more representative aggregated features. The aggregated features are input into a variational autoencoder and reparameterized to determine latent variables. The variational autoencoder can learn the latent distribution of the data, and reparameterization solves the differentiation problem in the sampling process, facilitating model training and optimization. The resulting latent variables contain key information from the original data, providing high-quality basic data for feature reconstruction and decoding prediction. This enables more accurate capture of abnormal patterns in the data during electricity fee anomaly detection, improving the accuracy and reliability of anomaly detection.

[0080] Preferably, in the embodiment, the feature reconstruction result includes reconstructed features, the electricity bill multiplier prediction result includes an electricity bill prediction multiplier, and the step of obtaining an electricity bill anomaly score based on the feature reconstruction result and the electricity bill multiplier prediction result includes: The reconstructed features are compared with the original features corresponding to the working parameters to obtain the feature reconstruction error; The predicted electricity price multiplier is compared with the actual electricity price multiplier data to obtain the electricity price multiplier deviation. The electricity bill anomaly score is obtained based on the feature reconstruction error and the electricity bill multiplier deviation.

[0081] Among them, the reconstructed features are the features that the feature reconstruction decoder attempts to restore to correspond to the base station's operating parameters, using latent variables as input. It reflects the model's ability to learn and reproduce the original operating parameters, and by comparing it with the original features, the accuracy of the model's simulation of the base station's operating state can be observed.

[0082] The electricity bill prediction ratio is a value derived by the electricity bill prediction decoder based on latent variables. This value is used to compare with the actual electricity bill prediction ratio data to determine the accuracy of the prediction and whether there are any anomalies.

[0083] Feature reconstruction error is the difference obtained by comparing the reconstructed features with the original features corresponding to the operating parameters. This error reflects the degree of deviation of the model when reconstructing the features of the operating parameters. The larger the error, the worse the model's restoration of the original features, which may indicate that there are abnormalities in the base station operating parameters or that the model's learning ability is insufficient.

[0084] The electricity tariff multiplier deviation is the difference between the predicted electricity tariff multiplier and the actual electricity tariff multiplier. It reflects the deviation between the predicted and actual electricity tariff multipliers. A large deviation may indicate problems with the configuration, metering, or prediction model of the electricity tariff multiplier.

[0085] The electricity bill anomaly score is a quantitative indicator derived from the combined feature reconstruction error and the electricity bill multiplier deviation, used to measure the likelihood of abnormal electricity bills at communication base stations. A higher score indicates a greater probability of abnormal electricity bills at the base station.

[0086] For example, the specific method for obtaining feature reconstruction error can be as follows: For each communication base station, the reconstructed features output by the feature reconstruction decoder are compared element-by-element or as a whole with the original features of the base station's operating parameters initially acquired. For example, if the power value in the original features is 50 kilowatts and the power value in the reconstructed features is 45 kilowatts, then the difference of 5 kilowatts is the feature reconstruction error, which quantifies the model's error in restoring the operating parameter features.

[0087] For example, the specific method for obtaining the electricity tariff deviation can be as follows: For each base station, the predicted electricity rate multiplier from the electricity rate multiplier is subtracted from the actual electricity rate multiplier. For example, if the actual electricity rate multiplier is 1.1 and the predicted multiplier is 1.3, then the difference of 0.2 is the electricity rate multiplier deviation, which measures the degree of deviation between the predicted and actual multipliers.

[0088] For example, the specific methods for obtaining an anomaly score in electricity charges can be as follows: By using weighted summation and function mapping, the feature reconstruction error and the electricity charge rate deviation are integrated into a single value, namely the electricity charge anomaly score. For example, assuming the weight of the feature reconstruction error is 0.6 and the weight of the electricity charge rate deviation is 0.4, if the feature reconstruction error of a certain base station is 3 and the electricity charge rate deviation is 0.5, then the electricity charge anomaly score = 3 × 0.6 + 0.5 × 0.4 = 1.8 + 0.2 = 2.

[0089] In this embodiment, the feature reconstruction error (comparing the reconstructed features with the original features) and the electricity charge multiplier deviation (comparing the predicted electricity charge multiplier with the actual electricity charge multiplier data) are obtained separately, and an electricity charge anomaly score is derived based on these two deviations. The feature reconstruction error reflects the accuracy of the model's learning of the base station's operating parameters. A large error may indicate an abnormality in the base station's operating status or problems in the data acquisition and processing process. The electricity charge multiplier deviation directly reflects the difference between the predicted and actual electricity charge multipliers, effectively identifying anomalies such as incorrect electricity charge multiplier configuration. The combined calculation of these two deviations yields the electricity charge anomaly score, considering both operating parameters and the electricity charge multiplier as key dimensions. This makes the measurement of base station electricity charge anomalies more comprehensive and accurate, enabling more precise location of abnormal base stations and effectively improving the sensitivity of electricity charge anomaly detection.

[0090] Preferably, in an embodiment, obtaining the electricity bill anomaly score based on the feature reconstruction error and the electricity bill multiplier deviation includes: Based on the historical false alarm rate or the cost of detecting abnormal electricity charges, determine the weighting coefficients corresponding to the feature reconstruction error and the deviation of the electricity charge multiplier. Based on the weighting coefficients, the feature reconstruction error and the electricity rate deviation are weighted and summed to obtain the basic anomaly score; Based on the potential distribution parameters, determine the potential distribution distance term or confidence term; The basic anomaly score is fused with the potential distribution distance term or the confidence term to obtain the electricity bill anomaly score.

[0091] The weighting coefficient, determined based on historical false alarm rates or electricity bill anomaly detection costs, is used to balance the relative importance of feature reconstruction errors and electricity bill multiplier deviations in calculating the electricity bill anomaly score. It can adjust the contribution ratio of these two deviations to the final anomaly score according to actual circumstances, making the anomaly score more closely reflect real-world needs.

[0092] The basic anomaly score is obtained by multiplying the feature reconstruction error and the electricity rate deviation by their respective weighting coefficients and then summing the results. It initially integrates the impact of feature reconstruction error and electricity rate deviation on anomalies, laying the foundation for further determining the final electricity anomaly score.

[0093] The latent distribution distance term is a value calculated based on the latent distribution parameters, reflecting the distance between the latent distribution of the current data and the normal distribution. This distance helps determine the degree to which the current data deviates from the normal state; the larger the distance, the more likely the data is to be in an abnormal distribution.

[0094] The confidence level is a numerical value determined based on the latent distribution parameters, representing the degree of confidence that the current data is normal or abnormal. The higher the confidence level, the more reliable the judgment on whether the data is abnormal.

[0095] For example, the weighting coefficients can be obtained in the following ways: Analyzing historical false alarm data reveals that a large number of false alarms are caused by feature reconstruction errors, indicating that these errors significantly interfere with anomaly detection, and their weighting coefficient may be appropriately reduced. Conversely, if the cost of detecting electricity bill anomalies is mainly related to the deviation in electricity bill multiplier, and the cost of correcting the multiplier deviation error is high, then the weighting coefficient of the electricity bill multiplier deviation should be increased accordingly. In this way, the weighting coefficient can be flexibly adjusted based on the actual situation.

[0096] For example, the basic anomaly score can be obtained in the following ways: The calculated feature reconstruction error is multiplied by its corresponding weighting coefficient, and the electricity rate deviation is multiplied by its corresponding weighting coefficient. The two products are then added together to obtain the basic anomaly score. For example, if the feature reconstruction error is 5 and its weighting coefficient is 0.6, and the electricity rate deviation is 0.3 and its weighting coefficient is 0.4, then the basic anomaly score = 5 × 0.6 + 0.3 × 0.4 = 3 + 0.12 = 3.12.

[0097] For example, the latent distribution distance term or confidence term can be obtained in the following ways: Using latent distribution parameters, the latent distribution distance term is calculated using a pre-defined algorithm (such as calculating the Euclidean distance to a known normal distribution); or a model is built based on the latent distribution parameters to output a confidence term indicating whether the current data is normal or abnormal. For example, by calculating the difference between the current latent distribution and the normal distribution in parameters such as mean and variance, the latent distribution distance term is found to be 2.5; the model predicts a confidence term of 0.8 that the current data is abnormal.

[0098] For example, the specific methods for obtaining an anomaly score in electricity charges can be as follows: The basic anomaly score is combined with the potential distribution distance term or confidence term using methods such as addition and weighted combination to obtain the final electricity bill anomaly score. For example, the basic anomaly score of 3.12 is directly added to the potential distribution distance term of 2.5, resulting in an electricity bill anomaly score of 3.12 + 2.5 = 5.62; or a weighted combination is used, such as a basic anomaly score weight of 0.7 and a potential distribution distance term weight of 0.3, resulting in an electricity bill anomaly score of 3.12 × 0.7 + 2.5 × 0.3 = 2.184 + 0.75 = 2.934.

[0099] In this embodiment, weighting coefficients are determined based on historical false alarm rates or electricity bill anomaly detection costs. This allows for flexible adjustment of the importance of feature reconstruction errors and electricity bill ratio deviations in anomaly score calculation, making the anomaly score more aligned with actual needs and reducing false alarms or missed alarms caused by fixed weights. A weighted summation is used to obtain the basic anomaly score, initially integrating the impact of two key deviations on anomalies by considering both base station operating parameters and electricity bill ratios, thus improving the accuracy of anomaly detection. Furthermore, determining the potential distribution distance or confidence term based on potential distribution parameters and integrating it with the basic anomaly score further introduces information at the potential distribution level, providing an additional reference dimension for anomaly judgment. This ensures that the final electricity bill anomaly score not only considers direct deviations but also incorporates the relationship between the potential distribution and the normal distribution, as well as the credibility of the anomaly judgment, thereby more comprehensively and accurately reflecting the degree of base station electricity bill anomalies and significantly improving the sensitivity and accuracy of electricity bill anomaly detection.

[0100] Preferably, in this embodiment, determining the abnormal electricity bill base station and the abnormal electricity bill type based on the abnormal electricity bill score includes: Obtain the preset threshold for abnormal electricity bill scores; The abnormal electricity charges score of each communication base station is compared with the abnormal electricity charges score threshold, and communication base stations with an abnormal electricity charges score greater than the abnormal electricity charges score threshold are selected as the abnormal electricity charges base stations. Calculate the proportion of the feature reconstruction error and the electricity rate deviation corresponding to the abnormal electricity charges base station in the abnormal electricity charge score; The type of electricity bill anomaly is determined based on the stated percentage.

[0101] The preset electricity bill anomaly score threshold is a numerical standard pre-set based on historical data, business experience, or specific detection requirements before conducting electricity bill anomaly detection. This threshold serves as the dividing line for judging whether the electricity bill of a communication base station is abnormal, distinguishing between normal and abnormal situations.

[0102] Among them, the abnormal electricity bill base station refers to the base station that is judged to have a high probability of abnormal electricity bill after comparing the abnormal electricity bill score of each communication base station with the preset abnormal electricity bill score threshold, that is, the base station whose abnormal electricity bill score is greater than the preset threshold.

[0103] The feature reconstruction error ratio refers to the proportion of feature reconstruction errors in the electricity charge anomaly score corresponding to a base station with abnormal electricity charges. By calculating this ratio, we can understand the contribution of feature reconstruction issues related to the base station's operating parameters to the overall anomaly score in the case of the base station's anomaly.

[0104] The electricity charge ratio deviation percentage refers to the proportion of the electricity charge ratio deviation in the electricity charge anomaly score corresponding to the base station with abnormal electricity charges. This percentage reflects the magnitude of the impact of electricity charge ratio-related issues on the overall anomaly score in the case of the base station's anomaly.

[0105] The electricity bill anomaly type is determined by the relative proportions of feature reconstruction error and electricity bill rate deviation, which define the specific category of communication base station electricity bill anomalies. Different proportions correspond to different causes of anomalies. For example, a large proportion of feature reconstruction error may indicate that the anomaly is related to factors such as the base station's operating status and data collection; a large proportion of electricity bill rate deviation may suggest that the anomaly is mainly due to issues with electricity bill rate configuration and metering.

[0106] For example, the preset threshold for abnormal electricity bill scores can be obtained as follows: Based on the analysis of a large amount of historical communication base station electricity cost data, the threshold is set in combination with business needs and detection objectives. For example, by statistically analyzing the distribution of abnormal electricity cost scores of normal and abnormal base stations in historical data, a value that can better distinguish between normal and abnormal base stations is selected as the threshold; or the threshold is determined through multiple experiments and adjustments based on the acceptable level of false alarm rate and false negative rate for actual business operations.

[0107] For example, the specific method for filtering base stations with abnormal electricity charges can be as follows: The calculated electricity charge anomaly score for each communication base station is compared one by one with a preset electricity charge anomaly score threshold. If the electricity charge anomaly score of a base station is greater than the threshold, the base station is marked as an electricity charge anomaly base station; if it is less than or equal to the threshold, the electricity charge situation of the base station is considered normal. Through this simple and direct comparison operation, base stations that may have electricity charge anomalies can be quickly screened out.

[0108] For example, the percentage can be obtained in the following ways: For base stations identified as having abnormal electricity charges, the feature reconstruction error is divided by the abnormal electricity charge score to obtain the feature reconstruction error percentage; the electricity charge multiplier deviation is divided by the abnormal electricity charge score to obtain the electricity charge multiplier deviation percentage. For example, if a base station with abnormal electricity charges has a feature reconstruction error of 5 and an abnormal electricity charge score of 10, then the feature reconstruction error percentage is 5 ÷ 10 = 0.5; if the electricity charge multiplier deviation is 2 and the abnormal electricity charge score is 10, then the electricity charge multiplier deviation percentage is 2 ÷ 10 = 0.2.

[0109] For example, the specific methods for determining the type of abnormal electricity bill can be as follows: Compare the proportion of feature reconstruction error with the proportion of electricity rate deviation. If the proportion of feature reconstruction error is greater than the proportion of electricity rate deviation, it indicates that factors related to the base station's operating parameters may play a dominant role in the abnormal situation of the base station, and it can be preliminarily determined that the anomaly type is related to the base station's operating status, data collection, etc. Conversely, if the proportion of electricity rate deviation is greater than the proportion of feature reconstruction error, it indicates that factors related to the electricity rate may be the main cause of the anomaly, such as incorrect electricity rate configuration or metering device failure.

[0110] In this embodiment, by setting a preset threshold for electricity fee anomaly scores and comparing the scores of each base station with this threshold, base stations potentially exhibiting electricity fee anomalies can be quickly and effectively screened, narrowing the scope of anomaly investigation and improving detection efficiency. Further calculation of the proportions of feature reconstruction errors and electricity fee ratio deviations in the electricity fee anomaly scores, and determination of the anomaly type based on these proportions, allows for a more in-depth and detailed analysis of anomalies. This approach not only determines whether a base station has electricity fee anomalies but also preliminarily identifies the root cause of the anomaly—whether it is related to base station operating parameters or the electricity fee ratio—providing a clear direction for targeted solutions to electricity fee anomaly problems. This improves the accuracy and practicality of the entire electricity fee anomaly detection system, effectively avoiding blind investigations and saving human and material resources.

[0111] Preferably, in this embodiment, obtaining the preset electricity bill anomaly score threshold includes: Obtain historical base station electricity fee detection data and manual detection work order data to obtain a historical sample dataset; Based on the abnormal score distribution in the historical sample dataset, determine the initial score threshold corresponding to the specified quantile; The initial score threshold is adjusted based on the recall rate and detection accuracy constraints of the electricity bill anomaly detection. The adjusted threshold will be used as the preset threshold for abnormal electricity charges.

[0112] Among them, historical base station electricity cost detection data includes various data collected from past electricity cost detection of communication base stations, covering base station operating parameters, electricity cost multiplier data, and anomaly scores calculated from them, reflecting the electricity cost-related status of base stations at different points in time.

[0113] Among them, the manual inspection work order data is the work order information recorded after manual on-site inspection or verification of the base station. It clarifies whether the base station actually has any abnormal electricity charges and the specific description of the abnormality, providing a reliable basis for judging the true status of the base station's electricity charges.

[0114] Among them, the abnormal score distribution refers to the distribution pattern of abnormal electricity scores of all communication base stations in the historical sample dataset, such as the frequency distribution of scores in different intervals, which shows the overall characteristics of abnormal scores.

[0115] The specified quantile is a specific positional indicator selected within the outlier score distribution, used to determine the initial score threshold. Common examples include the 90th quantile and the 95th quantile, which indicate that a corresponding proportion of all outlier scores are lower than the value corresponding to that quantile.

[0116] The initial score threshold is a preliminary score limit for judging abnormal electricity charges, determined based on a specified quantile in the abnormal score distribution. It is the starting point for subsequent threshold adjustments.

[0117] In electricity bill anomaly detection, recall rate refers to the proportion of base stations with actual electricity bill anomalies that are correctly detected. A higher recall rate indicates that the detection method can discover more genuinely abnormal base stations.

[0118] Detection accuracy refers to the proportion of base stations that are actually abnormal among all those detected as abnormal. Higher detection accuracy indicates a larger proportion of truly abnormal base stations in the detection results and fewer false alarms.

[0119] For example, the specific methods for obtaining historical sample datasets can be as follows: Historical electricity cost monitoring data for communication base stations, along with corresponding manual inspection work order data, are collected from data storage systems, databases, or record documents. This data is then organized and integrated to ensure consistency and integrity, forming a historical sample dataset to prepare for subsequent analysis.

[0120] For example, the initial score threshold can be obtained in the following way: Perform statistical analysis on outlier scores in the historical sample dataset and plot the distribution of outlier scores (e.g., histogram, cumulative distribution function plot, etc.). Based on the analysis objectives and experience, select an appropriate quantile (e.g., the 95th quantile). Determine the specific score value corresponding to this quantile by calculation or by consulting the distribution chart, and use this as the initial score threshold. This threshold initially delineates the boundary between normal and outlier scores.

[0121] For example, the initial score threshold can be adjusted in the following ways: Anomaly detection simulations were performed on base stations in a historical sample dataset using an initial scoring threshold. The number of detected anomalous base stations and the number of actually anomalous base stations among them were counted, thereby calculating the recall and precision. If the recall rate was lower than expected, it indicated that a large number of actual anomalous base stations might have gone undetected. In this case, the initial scoring threshold should be appropriately lowered to allow more base stations to be identified as anomalous. If the precision was low, i.e., there were many false positives, the initial scoring threshold should be increased to reduce the number of base stations identified as anomalous. By continuously adjusting the thresholds and recalculating the recall and precision, the pre-set recall and precision constraints were met.

[0122] After the above adjustments, when recall and detection accuracy reach a satisfactory balance and meet business requirements, this threshold will be determined as the preset electricity charge anomaly score threshold. This final threshold will be used for actual communication base station electricity charge anomaly detection as a standard for judging whether a base station is abnormal.

[0123] In this embodiment, a historical sample dataset is constructed by acquiring historical base station electricity fee detection data and manual detection work order data. An initial score threshold is determined using the anomaly score distribution within this dataset. Based on historical experience, a preliminary anomaly judgment standard with some reference value is established, reflecting the anomaly distribution characteristics in past detections. Then, the initial score threshold is adjusted according to the recall rate and detection accuracy constraints of electricity fee anomaly detection, ensuring that the threshold adapts to actual business needs and balances false positives and false negatives in the detection process. On the one hand, by reasonably adjusting the threshold to improve the recall rate, more base stations with actual electricity fee anomalies can be detected, avoiding false negatives; on the other hand, considering detection accuracy can effectively reduce false positives, making the detection results more reliable. The final preset electricity fee anomaly score threshold, which integrates historical data and business requirements, can more accurately filter out truly abnormal electricity fee base stations, thereby improving the accuracy and reliability of the entire electricity fee anomaly detection system, optimizing the anomaly detection process, and improving detection efficiency and quality.

[0124] To better implement the above methods, this application also provides a communication base station electricity fee anomaly detection device. This communication base station electricity fee anomaly detection device can be integrated into electronic devices, such as servers or terminals. The terminal may include tablet computers, laptop computers, and / or personal computers.

[0125] For example, see Figure 2 , Figure 2 This is a schematic diagram of the structure of the communication base station electricity fee anomaly detection device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the electricity fee anomaly detection device for this communication base station may include an acquisition module 201, a construction module 202, an aggregation module 203, a decoding module 204, and a detection module 205, as follows: The acquisition module 201 is used to acquire the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area; Construction module 202 is used to construct a base station association diagram based on the base station working service logic data; The aggregation module 203 is used to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph through a graph attention network to obtain aggregation features, and determine potential variables based on the aggregation features; Decoding module 204 is used to input the latent variables into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result; The detection module 205 is used to obtain an electricity fee anomaly score based on the feature reconstruction result and the electricity fee multiplier prediction result, and to determine the electricity fee anomaly base station and the electricity fee anomaly type based on the electricity fee anomaly score.

[0126] In practice, the above modules can be implemented as independent entities or combined in any way to be implemented as the same or several entities. For the specific implementation of the above modules, please refer to the previous method implementation examples, which will not be repeated here.

[0127] As can be seen from the above, in this embodiment of the application, a base station association graph is constructed based on the base station working service logic data. The base station association graph can break the limitation of treating base stations as independent individuals, utilize the relationships between base stations to mine potential anomalies, and improve the sensitivity to anomalies. By using a graph attention network to weight and aggregate various types of data and determine potential variables, potential connections and features between data can be mined. The potential variables are input into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction results and the electricity rate prediction results, which are analyzed from two dimensions. Based on these two results, an electricity rate anomaly score is obtained, and the base station and type of electricity rate anomaly are determined accordingly. Through multi-dimensional analysis and comprehensive judgment, electricity rate anomalies such as misconfiguration can be identified more comprehensively and accurately, thereby improving the sensitivity to these anomalies, reducing false alarms, and improving detection efficiency.

[0128] In some embodiments, the electricity fee anomaly detection device for a communication base station of the present invention can be implemented in a combination of hardware and software. As an example, the electricity fee anomaly detection device for a communication base station of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the electricity fee anomaly detection method for a communication base station of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0129] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0130] An electronic device according to an embodiment of the present invention includes a processor and a memory, wherein the memory stores multiple instructions; the processor loads instructions from the memory to execute the steps in the communication base station electricity fee anomaly detection method described above. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the communication base station electricity fee anomaly detection method shown in any embodiment of the present invention by calling the computer program.

[0131] In one alternative embodiment, an electronic device is provided, such as Figure 3 As shown, Figure 3 The illustrated electronic device 3000 includes a processor 3001 and a memory 3003. The processor 3001 and the memory 3003 are connected, for example, via a bus 3002. Optionally, the electronic device 3000 may further include a transceiver 3004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 3004 is not limited to one type, and the structure of the electronic device 3000 does not constitute a limitation on the embodiments of the present invention.

[0132] Processor 3001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 3001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0133] Bus 3002 may include a path for transmitting information between the aforementioned components. Bus 3002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 3002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The bus 3002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0134] The memory 3003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0135] The memory 3003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 3001. The processor 3001 executes the application code stored in the memory 3003 to implement the content shown in the foregoing method embodiments.

[0136] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0137] It should be noted that, Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0138] An embodiment of the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute the steps in the communication base station electricity fee anomaly detection method described above.

[0139] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0140] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the steps in the electricity bill anomaly detection method for a communication base station described above.

[0141] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0142] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the implementable architecture, functions, and operations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0143] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0144] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0145] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0146] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0147] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0148] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting abnormal electricity charges in a communication base station, characterized in that, include: Acquire the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area; Based on the base station operational service logic data, a base station association diagram is constructed; The working parameters, the electricity rate data, and the base station association graph are weighted and aggregated using a graph attention network to obtain aggregated features, and latent variables are determined based on the aggregated features. The latent variables are input into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result; Based on the feature reconstruction results and the electricity cost ratio prediction results, an electricity cost anomaly score is obtained, and based on the electricity cost anomaly score, the base station with electricity cost anomaly and the type of electricity cost anomaly are determined.

2. The method for detecting abnormal electricity charges in a communication base station according to claim 1, characterized in that, The base station operational service logic data includes station affiliation, power supply affiliation, and geographical proximity. The step of constructing a base station association graph based on the base station operational service logic data includes: Each of the aforementioned communication base stations is treated as a node; Based on the station affiliation relationship, an undirected edge is established for the nodes corresponding to two communication base stations with the same station identifier; Based on the power supply attribution relationship, an undirected edge is established for nodes corresponding to two communication base stations that are at the same power supply source; Based on the geographical proximity relationship, an undirected edge is established for nodes corresponding to two communication base stations whose geographical distance is less than a preset distance threshold; The base station association graph is determined based on the nodes and the undirected edges.

3. The method for detecting abnormal electricity charges in a communication base station according to claim 1, characterized in that, The process involves using a graph attention network to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph to obtain aggregated features. Based on these aggregated features, latent variables are determined, including: Based on the operating parameters and the electricity rate data, construct a node feature matrix; Based on the base station association graph, determine the node adjacency matrix; The node feature matrix and the node adjacency matrix are transmitted to the graph attention network to perform weighted aggregation of the contributions of the multiple communication base stations through attention coefficients to obtain the aggregated features, wherein the attention coefficients are obtained by nonlinear transformation and normalization of the node feature matrix; The aggregated features are input into a variational autoencoder to encode the latent distribution parameters. The latent distribution parameters are reparameterized to obtain the latent variables.

4. The method for detecting abnormal electricity charges in a communication base station according to claim 3, characterized in that, The feature reconstruction result includes reconstructed features, the electricity bill multiplier prediction result includes an electricity bill prediction multiplier, and the electricity bill anomaly score is obtained based on the feature reconstruction result and the electricity bill multiplier prediction result, including: The reconstructed features are compared with the original features corresponding to the working parameters to obtain the feature reconstruction error; The predicted electricity price multiplier is compared with the actual electricity price multiplier data to obtain the electricity price multiplier deviation. The electricity bill anomaly score is obtained based on the feature reconstruction error and the electricity bill multiplier deviation.

5. The method for detecting abnormal electricity charges in a communication base station according to claim 4, characterized in that, The step of obtaining the electricity bill anomaly score based on the feature reconstruction error and the electricity bill multiplier deviation includes: Based on the historical false alarm rate or the cost of detecting abnormal electricity charges, determine the weighting coefficients corresponding to the feature reconstruction error and the deviation of the electricity charge multiplier. Based on the weighting coefficients, the feature reconstruction error and the electricity rate deviation are weighted and summed to obtain the basic anomaly score; Based on the potential distribution parameters, determine the potential distribution distance term or confidence term; The basic anomaly score is fused with the potential distribution distance term or the confidence term to obtain the electricity bill anomaly score.

6. The method for detecting abnormal electricity charges in a communication base station according to claim 4, characterized in that, The step of determining the abnormal electricity charges base station and the type of abnormal electricity charges based on the abnormal electricity charge score includes: Obtain the preset threshold for abnormal electricity bill scores; The abnormal electricity charges score of each communication base station is compared with the abnormal electricity charges score threshold, and communication base stations with an abnormal electricity charges score greater than the abnormal electricity charges score threshold are selected as the abnormal electricity charges base stations. Calculate the proportion of the feature reconstruction error and the electricity rate deviation corresponding to the abnormal electricity charges base station in the abnormal electricity charge score; The type of electricity bill anomaly is determined based on the stated percentage.

7. The method for detecting abnormal electricity charges in a communication base station according to claim 6, characterized in that, The process of obtaining the preset threshold for abnormal electricity charges includes: Obtain historical base station electricity fee detection data and manual detection work order data to obtain a historical sample dataset; Based on the abnormal score distribution in the historical sample dataset, determine the initial score threshold corresponding to the specified quantile; The initial score threshold is adjusted based on the recall rate and detection accuracy constraints of the electricity bill anomaly detection. The adjusted threshold will be used as the preset threshold for abnormal electricity charges.

8. A device for detecting abnormal electricity charges in a communication base station, characterized in that, include: The acquisition module is used to acquire the operating parameters, electricity rate data, and base station operating business logic data of multiple communication base stations within the target detection area. The construction module is used to construct a base station association diagram based on the base station working service logic data; The aggregation module is used to perform weighted aggregation of the operating parameters, the electricity rate data, and the base station association graph through a graph attention network to obtain aggregated features, and to determine potential variables based on the aggregated features; The decoding module is used to input the latent variables into the feature reconstruction decoder and the electricity rate prediction decoder respectively to obtain the feature reconstruction result and the electricity rate prediction result; The detection module is used to obtain an electricity charge anomaly score based on the feature reconstruction result and the electricity charge multiplier prediction result, and to determine the electricity charge anomaly base station and the electricity charge anomaly type based on the electricity charge anomaly score.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing multiple instructions; the processor loads instructions from the memory to execute the steps in the electricity fee anomaly detection method for a communication base station as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to execute the steps of the electricity fee anomaly detection method for a communication base station according to any one of claims 1 to 7.

Images