Abnormal detection method, device and equipment for wiring relation of electric energy metering equipment

Abstract

The invention relates to an abnormal detection method, device and equipment for the wiring relation of electric energy metering equipment. The method comprises the steps that a wiring relation undirected graph of the wiring relation of the electric energy metering equipment in a to-be-detected area is obtained, the wiring relation undirected graph comprises node features and edge features, the node features and the edge features are input into a pre-constructed feature processing model, the node features and the edge features are converted into features of the same semantic dimension, and the feature processing model is used for processing the wiring relation of the electric energy metering equipment in the to-be-detected area. And the converted node features and edge features are fused to obtain fusion features, then the fusion features are input into an anomaly detection model, and an anomaly detection result of the wiring relation of the electric energy metering equipment in the to-be-detected area is obtained through the anomaly detection model, the anomaly detection model is obtained through sample fusion features of the sample electric energy metering equipment wiring relation and early stop strategy training. By adopting the method, the reliability of an anomaly detection result is improved, and meanwhile, the detection efficiency of anomaly detection is also improved.

Description

Technical Field

[0001] The present application relates to the technical field of anomaly detection, and particularly to an anomaly detection method, device, computer device, computer-readable storage medium, and computer program product for the wiring relationship of electric energy metering equipment. Background Art

[0002] With the rapid advancement of the construction of smart grids and the wide deployment of power consumption information collection systems, the number and types of electric energy metering equipment are increasing day by day, and their wiring methods are becoming increasingly complex. Electric energy metering equipment mainly includes core devices such as electric energy meters, acquisition terminals, and junction boxes. The wiring standardization between these devices is directly related to the accuracy of metering data and power consumption safety. Currently, hundreds of millions of electric energy metering equipment need to be installed, maintained, and inspected every year, and the wiring standardization detection is a basic but extremely challenging task.

[0003] However, the current detection methods for detecting the wiring standardization of wiring relationships have the problem of low detection efficiency. Summary of the Invention

[0004] Based on this, it is necessary to provide an anomaly detection method, device, computer device, computer-readable storage medium, and computer program product for the wiring relationship of electric energy metering equipment that can improve the detection efficiency in view of the above technical problems.

[0005] In a first aspect, the present application provides an anomaly detection method for the wiring relationship of electric energy metering equipment, including:

[0006] Obtain an undirected graph of the wiring relationship of electric energy metering equipment in the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, the node features are features used to characterize the electric energy metering equipment, and the edge features are features used to characterize the connection relationship between the electric energy metering equipment and circuit elements;

[0007] Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features;

[0008] Input the fused features into an anomaly detection model, and obtain an anomaly detection result of the wiring relationship of electric energy metering equipment in the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the wiring relationship of sample electric energy metering equipment and an early stopping strategy.

[0009] In one of the embodiments, the feature processing model includes a hierarchical feature extraction module, and the hierarchical feature extraction module includes a first hierarchical feature extraction unit, a second hierarchical feature extraction unit, and a third hierarchical feature extraction unit;

[0010] Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features, including:

[0011] Input the node features and edge features into the first hierarchical feature extraction unit to obtain the node features and edge features after the first extraction;

[0012] Input the node features and edge features after the first extraction into the second hierarchical feature extraction unit to obtain the node features and edge features after the second extraction;

[0013] Input the node features and edge features after the second extraction into the third hierarchical feature extraction unit to obtain the node features and edge features after the third extraction; the corresponding feature extraction dimensions of the first hierarchical feature extraction unit, the second hierarchical feature extraction unit, and the third hierarchical feature extraction unit decrease in sequence;

[0014] Convert the node features and edge features after the third extraction into features in the same semantic dimension, and fuse the converted node features and edge features after the third extraction to obtain fused features.

[0015] In one embodiment, the feature processing model further includes a feature fusion module;

[0016] Convert the node features and edge features after the third extraction into features in the same semantic dimension, and fuse the converted node features and edge features after the third extraction to obtain fused features, including:

[0017] Input the node features and edge features after the third extraction into the feature fusion module, perform a linear mapping on the node features and edge features after the third extraction to obtain feature vectors in the same semantic dimension, and input each feature vector into the corresponding attention head and perform a linear mapping to obtain fused features.

[0018] In one embodiment, before obtaining the undirected graph of the wiring relationship of the electric energy metering device in the area to be detected, the method includes:

[0019] Obtain the on-site image data in the area to be detected, and input the on-site image data into a pre-constructed visual feature extraction model to obtain the visual features corresponding to the electric energy metering device;

[0020] Obtain the technical parameter description text and physical attribute data of the electric energy metering device, and input the technical parameter description text into a pre-constructed semantic feature extraction module to obtain the semantic features corresponding to the electric energy metering device; the physical attribute data includes wiring connection type data, spatial distance data, and electrical characteristic data;

[0021] Construct an undirected graph of the wiring relationship corresponding to the power metering device wiring relationship based on visual features, semantic features, and physical attribute data.

[0022] In an exemplary embodiment, constructing an undirected graph of the wiring relationship corresponding to the power metering device wiring relationship based on visual features, semantic features, and physical attribute data includes:

[0023] Stitch the visual features and semantic features to obtain the original node features;

[0024] Perform feature dimensionality reduction on the original node features to obtain the node features;

[0025] Perform edge feature encoding on the physical attribute data to obtain the edge features;

[0026] Based on the node features and edge features, construct an undirected graph of the wiring relationship of the power metering device wiring relationship.

[0027] In one of the embodiments, the anomaly detection model is trained through the following steps:

[0028] Obtain the sample undirected graph of the wiring relationship of the sample power metering device wiring relationship within the sample area, and obtain the sample fusion features of the sample power metering device wiring relationship according to the sample undirected graph of the wiring relationship;

[0029] Based on the sample fusion features, construct a training data set, a validation data set, and a test data set respectively;

[0030] Use the training data set to adjust the model parameters of the to-be-trained anomaly detection model to obtain a candidate anomaly detection model;

[0031] According to the validation data set, adjust the hyperparameters of the candidate anomaly detection model to obtain a to-be-tested anomaly detection model;

[0032] Use the test data set to perform a performance test on the to-be-tested anomaly detection model. If the test result of the performance test indicates that the to-be-tested anomaly detection model passes the performance test, then determine the to-be-tested anomaly detection model as the anomaly detection model.

[0033] In one embodiment, the method further includes:

[0034] If the test result of the performance test indicates that the to-be-tested anomaly detection model fails the performance test, and the number of consecutive failures reaches the set tolerance count threshold, stop parameter adjustment and use the current to-be-tested anomaly detection model as the anomaly detection model.

[0035] In a second aspect, the present application further provides an anomaly detection device for the power metering device wiring relationship, including:

[0036] An undirected graph acquisition module, configured to acquire an undirected graph of the wiring relationship of the power metering device within the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, where the node features are features used to characterize the power metering device, and the edge features are features used to characterize the connection relationship between the power metering device and the circuit components;

[0037] A feature acquisition module, configured to input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features;

[0038] An anomaly detection module, configured to input the fused features into an anomaly detection model, and obtain an anomaly detection result of the wiring relationship of the power metering device within the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the sample wiring relationship of the power metering device and an early stopping strategy.

[0039] In a third aspect, the present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0040] Acquire an undirected graph of the wiring relationship of the power metering device within the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, where the node features are features used to characterize the power metering device, and the edge features are features used to characterize the connection relationship between the power metering device and the circuit components;

[0041] Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features;

[0042] Input the fused features into an anomaly detection model, and obtain an anomaly detection result of the wiring relationship of the power metering device within the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the sample wiring relationship of the power metering device and an early stopping strategy.

[0043] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0044] Acquire an undirected graph of the wiring relationship of the power metering device within the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, where the node features are features used to characterize the power metering device, and the edge features are features used to characterize the connection relationship between the power metering device and the circuit components;

[0045] Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features;

[0046] Input the fused features into an anomaly detection model, and obtain the anomaly detection result of the wiring relationship of the power metering devices in the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the wiring relationships of the sample power metering devices and an early stopping strategy.

[0047] In a fifth aspect, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the following steps:

[0048] Obtain an undirected graph of the wiring relationship of the power metering devices in the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, where the node features are features used to characterize the power metering devices, and the edge features are features used to characterize the connection relationship between the power metering devices and circuit elements;

[0049] Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features;

[0050] Input the fused features into an anomaly detection model, and obtain the anomaly detection result of the wiring relationship of the power metering devices in the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the wiring relationships of the sample power metering devices and an early stopping strategy.

[0051] The above abnormal detection method, device, computer device, computer-readable storage medium and computer program product for the wiring relationship of the electric energy metering device obtain an undirected graph of the wiring relationship of the electric energy metering device in the area to be detected. Among them, the undirected graph of the wiring relationship includes node features and edge features. The node features are used to characterize the features of the electric energy metering device, and the edge features are used to characterize the features of the connection relationship between the electric energy metering device and the circuit elements. Input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain fused features. Then input the fused features into an abnormal detection model, and obtain the abnormal detection result of the wiring relationship of the electric energy metering device in the area to be detected through the abnormal detection model. Among them, the abnormal detection model is trained through the sample fused features of the sample electric energy metering device wiring relationship and an early stopping strategy. By obtaining an undirected graph for characterizing the wiring relationship of the electric energy metering device, and through feature processing of the node features and edge features included in the undirected graph of the wiring relationship, the corresponding fused features are obtained, and the corresponding abnormal detection results are obtained according to the fused features and the pre-trained abnormal detection model. By introducing an undirected graph, the efficiency and accuracy of abnormal detection are improved, and subsequent various models are used for fully automatic data processing and abnormal detection, improving the reliability of the abnormal detection results and also improving the detection efficiency of abnormal detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0053] Figure 1 It is an application environment diagram of the abnormal detection method for the wiring relationship of the electric energy metering device in an embodiment;

[0054] Figure 2 It is a schematic flowchart of the abnormal detection method for the wiring relationship of the electric energy metering device in an embodiment;

[0055] Figure 3 It is a schematic flowchart of the abnormal detection method for the wiring relationship of the electric energy metering device in another embodiment;

[0056] Figure 4 It is a structural block diagram of the abnormal detection device for the wiring relationship of the electric energy metering device in an embodiment;

[0057] Figure 5 It is an internal structure diagram of a computer device in an embodiment. Specific implementation manner

[0058] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0059] The abnormal detection method for the wiring relationship of the electric energy metering device provided by the embodiment of the present application can be applied to, for example Figure 1 the application environment shown in the figure. Among them, the server 102 can obtain various parameters of the electric energy metering device in the area to be detected through the communication network, including on-site image data, technical parameter description text, and physical attribute data. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 102, or can be placed in the cloud or other network servers. Obtain the undirected graph of the wiring relationship of the electric energy metering device in the area to be detected from the data storage system, where the undirected graph of the wiring relationship includes node features and edge features, the node features are the features used for the electric energy metering device, and the edge features are the features used to represent the connection relationship between the electric energy metering device and the circuit element. Input the node features and edge features into the pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain a fused feature. Input the fused feature into the abnormal detection model, and obtain the abnormal detection result of the wiring relationship of the electric energy metering device in the area to be detected through the abnormal detection model. Among them, the abnormal detection model is trained through the sample fused features of the sample electric energy metering device wiring relationship and the early stopping strategy. The server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0060] In an exemplary embodiment, as Figure 2 shown, a method for detecting abnormal wiring relationship of an electric energy metering device is provided. Taking the method applied to Figure 1 the server 102 in the figure as an example, it includes the following steps S201 to S203. Among them:

[0061] Step S201, obtain the undirected graph of the wiring relationship of the electric energy metering device in the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, the node features are the features used to represent the electric energy metering device, and the edge features are the features used to represent the connection relationship between the electric energy metering device and the circuit element.

[0062] Among them, the electric energy metering device can be understood as an instrument used to measure the electric energy consumption and generation. The node feature can be understood as a feature used to characterize the electric energy metering device, and the edge feature can be understood as a feature used to characterize the connection relationship between the electric energy metering device and the circuit element. The undirected graph of the wiring relationship can be understood as an undirected image used to characterize the connection relationship in the circuit element.

[0063] Exemplarily, the server 102 obtains the on-site image data within the area to be detected, as well as the technical parameter description text and physical attribute data of the electric energy metering device. Based on the on-site image data and the technical parameter description text, the node features used to characterize the electric energy metering device are obtained, and based on the physical attribute data, the edge features used to characterize the connection relationship between the electric energy metering device and the circuit element are obtained. Finally, based on the node features and the edge features, the undirected graph of the wiring relationship of the electric energy metering device within the area to be detected is obtained. The undirected graph of the wiring relationship obtained in the above manner models the actual situation of the area to be detected more accurately, thereby improving the accuracy of the abnormal detection result of the wiring relationship.

[0064] Step S202: Input the node features and the edge features into a pre-constructed feature processing model, convert the node features and the edge features into features of the same semantic dimension, and fuse the converted node features and edge features to obtain fused features.

[0065] Optionally, the server 102 inputs the node features and the edge features into a pre-constructed feature processing model, first performs hierarchical feature extraction on the node features and the edge features to obtain the extracted node features and edge features, then projects the extracted node features and edge features into the same feature space through linear projection, and inputs the projected node features and edge features into multiple attention heads, and adds the outputs of each attention head and outputs the projection to obtain fused features. Through the above process of feature fusion, information from different feature sources can be quickly integrated, thereby optimizing the calculation efficiency and reducing redundancy.

[0066] Step S203: Input the fused features into an abnormal detection model, and obtain the abnormal detection result of the wiring relationship of the electric energy metering device within the area to be detected through the abnormal detection model; the abnormal detection model is trained through the sample fused features of the wiring relationship of the sample electric energy metering device and the early stopping strategy.

[0067] Among them, the anomaly detection model can be understood as a multi-layer perceptron. The multi-layer perceptron is trained through the sample fusion features of the wiring relationships of sample power metering devices and an early stopping strategy. The early stopping strategy can be understood as a regularization technique commonly used in machine learning and deep learning, mainly used to prevent the model from overfitting. Its basic idea is to monitor the performance of the model on the test set and stop training in advance when it is found that the performance fails the test continuously for multiple times; the anomaly detection results can be understood as different misclassifications and the occurrence probabilities of each misclassification.

[0068] Exemplarily, the server 102 inputs the fusion features into the anomaly detection model, and obtains the anomaly detection results of the wiring relationships of the power metering devices in the area to be detected through the anomaly detection model:

[0069] Based on the fused features, the final misclassification and localization are performed through the designed multi-layer perceptron:

[0070]

[0071] Among them, y ∈ R^C is the predicted probability distribution of error types, and C is the predefined number of error categories (in this embodiment, C = 12, including common fault types such as phase sequence errors and reverse polarity connections).

[0072] Through the combination of the fusion features and the multi-layer perceptron, the model can better capture the complex patterns in the wiring relationships, thereby improving the accuracy of anomaly detection and misclassification; the predefined fault categories enable the model not only to detect anomalies but also to clearly indicate the specific fault types, facilitating subsequent maintenance and correction; through anomaly detection, the server 102 can monitor the working status of the power metering devices in real time, quickly identify those wiring configurations or states that deviate from the normal operating mode, and issue alarms in a timely manner.

[0073] In the above abnormal detection method for the wiring relationship of power metering devices, an undirected graph of the wiring relationship of power metering devices in the area to be detected is obtained. Among them, the undirected graph of the wiring relationship includes node features and edge features. The node features are the features used to characterize the power metering devices, and the edge features are the features used to characterize the connection relationship between the power metering devices and circuit elements. The node features and edge features are input into a pre-constructed feature processing model, the node features and edge features are converted into features in the same semantic dimension, and the converted node features and edge features are fused to obtain fused features. Then, the fused features are input into an abnormal detection model, and an abnormal detection result of the wiring relationship of power metering devices in the area to be detected is obtained through the abnormal detection model. Among them, the abnormal detection model is trained through the sample fused features of the sample power metering device wiring relationship and an early stopping strategy. By obtaining an undirected graph of the wiring relationship used to characterize the power metering devices, and through feature processing of the node features and edge features included in the undirected graph of the wiring relationship, corresponding fused features are obtained, and corresponding abnormal detection results are obtained based on the fused features and a pre-trained abnormal detection model. By introducing an undirected graph, the efficiency and accuracy of abnormal detection are improved, and subsequent use of various models for fully automatic data processing and abnormal detection improves the reliability of the abnormal detection results and also improves the detection efficiency of abnormal detection.

[0074] In one embodiment, the feature processing model includes a hierarchical feature extraction module, and the hierarchical feature extraction module includes a first hierarchical feature extraction unit, a second hierarchical feature extraction unit, and a third hierarchical feature extraction unit;

[0075] Inputting the node features and edge features into a pre-constructed feature processing model, converting the node features and edge features into features in the same semantic dimension, and fusing the converted node features and edge features to obtain fused features includes:

[0076] Inputting the node features and edge features into the first hierarchical feature extraction unit to obtain the node features and edge features after the first extraction; inputting the node features and edge features after the first extraction into the second hierarchical feature extraction unit to obtain the node features and edge features after the second extraction; inputting the node features and edge features after the second extraction into the third hierarchical feature extraction unit to obtain the node features and edge features after the third extraction; the corresponding feature extraction dimensions of the first hierarchical feature extraction unit, the second hierarchical feature extraction unit, and the third hierarchical feature extraction unit decrease in sequence; converting the node features and edge features after the third extraction into features in the same semantic dimension, and fusing the converted node features and edge features after the third extraction to obtain fused features.

[0077] Optionally, the server 102 adopts an improved Graph Convolutional Network (GCN) architecture to achieve in-depth updates of node features through multi-layer feature transfer and aggregation operations, so as to fully capture the local and global topological relationship information between devices.

[0078] Specifically, the feature update formula of the l-th layer of GCN is:

[0079]

[0080] where the definitions of each parameter are as follows: Adjacency matrix: , and self-connection is achieved by adding the identity matrix . Degree matrix: , which is used for feature normalization. Node feature matrix: , which contains the feature representations of all nodes in the l-th layer. Learnable weight: and bias term . Activation function: , and LeakyReLU (Leaky Rectified Linear Unit) is adopted to provide non-linear transformation ability.

[0081] In this implementation, 3 layers of GCN are set, and the feature dimensions are 512→256→128 respectively to extract higher-level abstract features layer by layer. The node features and edge features extracted layer by layer are converted into features in the same semantic dimension, and the converted node features and edge features extracted layer by layer are fused to obtain fused features.

[0082] The advantages of hierarchical feature extraction are as follows: expanding the receptive field layer by layer, realizing local-to-global feature extraction, extracting more abstract high-level features through dimensionality reduction operations, and maintaining controllability of computational complexity.

[0083] In one of the embodiments, the feature processing model further includes a feature fusion module; the node features and edge features extracted three times are converted into features in the same semantic dimension, and the converted node features and edge features extracted three times are fused to obtain fused features, including: inputting the node features and edge features extracted three times into the feature fusion module, performing linear mapping on the node features and edge features extracted three times to obtain feature vectors in the same semantic dimension, inputting each feature vector into the corresponding attention head and performing linear mapping to obtain fused features.

[0084] Exemplarily, in order to achieve in-depth fusion of visual, text, and graph structure features, the server 102 designs a multi-head attention mechanism based on Transformer. First, the features of different modalities are mapped to the same feature space through linear projection:

[0085]

[0086] Among them , is 64. Then calculate the multi-head attention:

[0087]

[0088] Among them, the number of attention heads h is 8, the scaling factor is used for gradient stability, and the output projection .

[0089] Through the above process, feature fusion is performed on the node features and edge features. The obtained fusion features can better represent the typical information of the cable tunnel, reduce the subsequent data processing volume, and at the same time improve the detection accuracy of the anomaly detection results.

[0090] In an exemplary embodiment, before obtaining the undirected graph of the wiring relationship of the power metering device in the area to be detected, the method includes:

[0091] Obtain the on-site image data in the area to be detected, and input the on-site image data into a pre-constructed visual feature extraction model to obtain the visual features corresponding to the power metering device; obtain the technical parameter description text and physical attribute data of the power metering device, and input the technical parameter description text into a pre-constructed semantic feature extraction module to obtain the semantic features corresponding to the power metering device; the physical attribute data includes wiring connection type data, spatial distance data, and electrical characteristic data; construct an undirected graph of the wiring relationship corresponding to the power metering device according to the visual features, semantic features, and physical attribute data.

[0092] Optionally, the server 102 extracts two types of features in parallel: visual features : Process the on-site image of the device through a pre-trained ResNet-50 deep convolutional neural network to extract 2048-dimensional high-level visual representations, which contain key visual information such as the appearance, installation status, and wiring position of the device; text features : Use the BERT pre-trained language model to process the technical parameter description text of the device to generate 768-dimensional semantic feature vectors, which effectively encode professional knowledge such as the specification parameters and model information of the device.

[0093] Extract the corresponding visual features through the on-site image data, extract the corresponding semantic features through the technical parameter description text of the power metering device. The multi-type features have better performance for the on-site situation. At the same time, obtain the physical attribute data of the power metering device. The undirected graph constructed according to the visual features, semantic features, and physical attribute data has better performance for the actual situation, and is also more accurate, thereby improving the detection results of the anomaly detection.

[0094] In one embodiment, according to visual features, semantic features, and physical attribute data, an undirected wiring relationship graph corresponding to the wiring relationship of the power metering device is constructed, including: splicing visual features and semantic features to obtain original node features; performing feature dimensionality reduction on the original node features to obtain node features; performing edge feature encoding on the physical attribute data to obtain edge features; and constructing an undirected wiring relationship graph of the power metering device wiring relationship based on the node features and edge features.

[0095] Exemplarily, subsequently, the server 102 constructs a unified node feature representation through a feature fusion operation:

[0096]

[0097] where [;] represents the feature splicing operation, and the initial feature dimension is 2816. To improve computational efficiency and retain key information, the features are reduced to 512 dimensions through a learnable linear transformation:

[0098]

[0099] For the extraction of edge features, a dedicated encoding function is designed: This function comprehensively considers the physical connection attributes between devices, including connection types (such as three-phase four-wire system, single-phase, etc.), spatial distance information, electrical characteristic parameters (voltage level, impedance, etc.), and finally generates a 64-dimensional edge feature vector ( = 64).

[0100]

[0101] where is the edge feature encoding function, is the edge feature dimension.

[0102] Based on the node features and edge features, the wiring relationship of the power metering device is modeled as an undirected graph structure , where represents various key device nodes in the system, including but not limited to core metering devices such as electricity meters, acquisition terminals, junction boxes, and transformers. The edge set , represents the physical connection relationship between devices, reflecting the actual topological structure of the system.

[0103] Through feature fusion and dimensionality reduction, the model can comprehensively utilize visual and text information to obtain corresponding node features. At the same time, the obtained node features have better representation ability. Secondly, the edge feature encoding function is used to encode the physical attribute data to obtain corresponding edge features. Based on the node features and edge features, a more representative and simpler-structured undirected graph can be constructed, thereby improving the detection efficiency and detection accuracy of anomaly detection.

[0104] In one embodiment, the anomaly detection model is trained through the following steps: obtaining an undirected graph of the sample wiring relationship of the sample power metering devices in the sample area, and obtaining the sample fusion features of the sample wiring relationship of the sample power metering devices according to the undirected graph of the sample wiring relationship; respectively constructing a training data set, a validation data set, and a test data set based on the sample fusion features; using the training data set to adjust the model parameters of the anomaly detection model to be trained to obtain a candidate anomaly detection model; according to the validation data set, adjusting the hyperparameters of the candidate anomaly detection model to obtain an anomaly detection model to be tested; using the test data set to perform a performance test on the anomaly detection model to be tested. If the test result of the performance test indicates that the anomaly detection model to be tested passes the performance test, then the anomaly detection model to be tested is determined as the anomaly detection model.

[0105] Optionally, the server 102 obtains an undirected graph of the sample wiring relationship of the sample power metering devices in the sample area, and obtains the sample fusion features of the sample wiring relationship of the sample power metering devices according to the undirected graph of the sample wiring relationship. Then, based on the sample fusion features, a training data set, a validation data set, and a test data set are respectively constructed. The training data set is used to adjust the model parameters of the anomaly detection model to be trained to obtain a candidate anomaly detection model: inputting the training data set into the anomaly detection model to be trained to obtain a predicted anomaly detection result corresponding to the training data set. According to the predicted anomaly detection result, the actual anomaly detection result, and the weighted cross-entropy loss function preset for the anomaly detection model to be trained, the corresponding model loss value is obtained. The model parameters of the anomaly detection model to be trained are adjusted according to the model loss value until the model loss value reaches the minimum value, and the current anomaly detection model to be trained is used as the candidate anomaly detection model.

[0106] According to the validation data set and the cosine annealing strategy, the hyperparameters of the candidate anomaly detection model are adjusted to obtain an anomaly detection model to be tested. The test data set is used to perform a performance test on the anomaly detection model to be tested. If the test result of the performance test indicates that the anomaly detection model to be tested passes the performance test, then the anomaly detection model to be tested is determined as the anomaly detection model.

[0107] The division and construction of the training, validation, and test data sets and their simultaneous utilization in the model training process ensure the comprehensiveness and effectiveness of model evaluation and help improve the generalization ability of the model. Secondly, the application of the weighted cross-entropy loss function and the cosine annealing strategy makes the model more efficient in the optimization process and can be quickly adjusted to the optimal state.

[0108] In one embodiment, the method further includes: if the test result of the performance test indicates that the anomaly detection model to be tested fails the performance test and the number of consecutive failures reaches the set tolerance threshold, stop adjusting the parameters and use the current anomaly detection model to be tested as the anomaly detection model.

[0109] Exemplarily, if the test result of the performance test indicates that the anomaly detection model to be tested fails the performance test and the number of consecutive failures reaches the set tolerance threshold, stop adjusting the model parameters and the model hyperparameters, and directly use the last current anomaly detection model to be tested as the anomaly detection model. By setting the tolerance threshold, overfitting caused by unnecessary parameter adjustment of the model is avoided, and the adjustment is stopped when the performance test fails, effectively saving time and computing resources.

[0110] In an exemplary embodiment, as Figure 3 shown, a specific implementation process of an anomaly detection method for the wiring relationship of an electric energy metering device is provided (the following data are all examples and do not limit that the solution of the present application can only be implemented in this case), where:

[0111] Step S1: Graph structure modeling and feature initialization:

[0112] To comprehensively characterize the features of each device node, this solution innovatively adopts a multi-modal feature extraction strategy. For each node , the system extracts two types of features in parallel: visual features : Process the on-site image of the device through a pre-trained ResNet-50 deep convolutional neural network to extract 2048-dimensional high-level visual representations, which contain key visual information such as the appearance, installation status, and wiring position of the device; text features : Use the BERT pre-trained language model to process the technical parameter description text of the device to generate a 768-dimensional semantic feature vector, which effectively encodes professional knowledge such as the specification parameters and model information of the device. Subsequently, a unified node feature representation is constructed through a feature fusion operation:

[0113]

[0114] where [;] represents the feature concatenation operation, and the initial feature dimension is 2816. To improve the calculation efficiency and retain key information, the feature is reduced to 512 dimensions through a learnable linear transformation:

[0115]

[0116] For the extraction of edge features, a dedicated encoding function is designed: This function comprehensively considers the physical connection attributes between devices, including connection types (such as three-phase four-wire system, single-phase, etc.), spatial distance information, and electrical characteristic parameters (voltage level, impedance, etc.), and finally generates a 64-dimensional edge feature vector ( = 64).

[0117]

[0118] Among them is the edge feature encoding function, is the edge feature dimension.

[0119] Based on the node features and edge features obtained above, the wiring relationship of the electric energy metering device is modeled as an undirected graph structure , where represents various key device nodes in the system, including but not limited to core metering devices such as electric energy meters, acquisition terminals, junction boxes, and current transformers. The edge set represents the physical connection relationship between devices and reflects the actual topological structure of the system.

[0120] Step S2: Hierarchical feature extraction of the graph neural network:

[0121] In this step, an improved graph convolutional network (GCN) architecture is adopted. Through multi-layer feature transfer and aggregation operations, the depth update of node features is realized to fully capture the local and global topological relationship information between devices.

[0122] Specifically, the feature update formula of the l-th layer of GCN is:

[0123]

[0124] The definitions of each parameter are as follows: Adjacency matrix: , realizing self-connection by adding the identity matrix . Degree matrix: , used for feature normalization. Node feature matrix: , containing the feature representations of all nodes in the l-th layer. Learnable weights: and bias term . Activation function: , using LeakyReLU to provide non-linear transformation ability.

[0125] In this implementation, 3 layers of GCN are set, and the feature dimensions are 512 → 256 → 128 respectively, to extract higher-level abstract features layer by layer. The advantages of this hierarchical feature extraction are: expanding the receptive field layer by layer, realizing feature extraction from local to global, extracting more abstract high-level features through dimensionality reduction operations, and maintaining the controllability of computational complexity.

[0126] Step S3: Multi-modal Feature Fusion Based on Transformer:

[0127] To achieve the deep fusion of visual, text, and graph structure features, a multi-head attention mechanism based on Transformer is designed. First, the features of different modalities are mapped to the same feature space through linear projection:

[0128]

[0129] where , is 64. Then, the multi-head attention is calculated:

[0130]

[0131] where the number of attention heads h is 8, the scaling factor is used for gradient stabilization, and the output projection .

[0132] Step S4: Error Classification and Localization Module:

[0133] Based on the fused features, the final error classification and localization are performed through a designed multi-layer perceptron:

[0134]

[0135] where y ∈ R^C is the predicted probability distribution of error types, and C is the predefined number of error categories (in this implementation, C = 12, including common fault types such as phase sequence error and reverse polarity connection).

[0136] Step S5: Loss Function Design and Model Training (Step S5 is before Step S4, or can be considered to be carried out synchronously with Step S4):

[0137] The cross-entropy loss function with weights is used to optimize the model:

[0138]

[0139] where N is the number of batch samples, represents all learnable parameters of the model, is the regularization coefficient 0.0001.

[0140] The Adam optimizer is used for training, the initial learning rate is set to 0.001, and the cosine annealing strategy is used for learning rate adjustment. To prevent overfitting, an early stopping strategy is introduced, and the training stops when the performance on the validation set has not improved for 5 consecutive epochs.

[0141] Compared with the prior art, the present application has the following advantages:

[0142] 1. By constructing a multi-modal deep learning framework based on a graph structure, the overall performance of the wiring normality detection of power metering devices has been improved.

[0143] 2. The above method significantly improves the detection accuracy, while greatly reducing the false alarm rate and missed alarm rate, enhancing the adaptability of the system to different models of devices and the detection stability in complex on-site environments, and improving the tolerance to data noise and device occlusion.

[0144] 3. It significantly shortens the single detection time, realizes batch parallel processing capabilities, and reduces the consumption of computing resources; it realizes the precise positioning of wiring errors, the identification and classification of multiple types of errors, and can provide detailed error analysis reports and correction suggestions.

[0145] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages, and these steps or stages do not necessarily have to be executed at the same time, but can be executed at different times, and the execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or in turn with at least some of the steps or stages in other steps or other steps.

[0146] Based on the same inventive concept, the embodiments of the present application also provide an abnormal detection device for the wiring relationship of power metering devices for implementing the abnormal detection method of the wiring relationship of power metering devices involved above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the abnormal detection device for the wiring relationship of power metering devices provided below can refer to the limitations on the abnormal detection method of the wiring relationship of power metering devices in the above text, and will not be repeated here.

[0147] In an exemplary embodiment, as Figure 4 shown, an abnormal detection device for the wiring relationship of power metering devices is provided, including: an undirected graph acquisition module 401, a feature acquisition module 402, and an abnormal detection module 403, where:

[0148] An undirected graph acquisition module 401 is used to acquire an undirected graph of the wiring relationship of the power metering equipment in the area to be detected; the undirected graph of the wiring relationship includes node features and edge features, where the node features are features used to characterize the power metering equipment, and the edge features are features used to characterize the connection relationship between the power metering equipment and the circuit components;

[0149] A feature acquisition module 402 is used to input the node features and edge features into a pre-constructed feature processing model, convert the node features and edge features into features in the same semantic dimension, and fuse the converted node features and edge features to obtain a fused feature;

[0150] An anomaly detection module 403 is used to input the fused feature into an anomaly detection model, and obtain an anomaly detection result of the wiring relationship of the power metering equipment in the area to be detected through the anomaly detection model; the anomaly detection model is trained through the sample fused features of the wiring relationship of the sample power metering equipment and an early stopping strategy.

[0151] In one embodiment, the feature processing model includes a hierarchical feature extraction module, and the hierarchical feature extraction module includes a first hierarchical feature extraction unit, a second hierarchical feature extraction unit, and a third hierarchical feature extraction unit; the feature acquisition module 402 further includes a primary extraction sub-module, a secondary extraction sub-module, a tertiary extraction sub-module, and a feature fusion sub-module, where:

[0152] The primary extraction sub-module is used to input the node features and edge features into the first hierarchical feature extraction unit to obtain the node features and edge features after primary extraction;

[0153] The secondary extraction sub-module is used to input the node features and edge features after primary extraction into the second hierarchical feature extraction unit to obtain the node features and edge features after secondary extraction;

[0154] The tertiary extraction sub-module is used to input the node features and edge features after secondary extraction into the third hierarchical feature extraction unit to obtain the node features and edge features after tertiary extraction; the feature extraction dimensions corresponding to the first hierarchical feature extraction unit, the second hierarchical feature extraction unit, and the third hierarchical feature extraction unit decrease in sequence;

[0155] The feature fusion sub-module is used to convert the node features and edge features after tertiary extraction into features in the same semantic dimension, and fuse the converted node features and edge features after tertiary extraction to obtain a fused feature.

[0156] In one embodiment, the feature processing model further includes a feature fusion module; the feature fusion sub-module is further configured to input the node features and edge features after three extractions into the feature fusion module, perform a linear mapping on the node features and edge features after three extractions to obtain feature vectors in the same semantic dimension, input each feature vector into the corresponding attention head and perform a linear mapping to obtain fused features.

[0157] In an exemplary embodiment, before obtaining the undirected graph of the wiring relationship of the power metering device in the area to be detected, the abnormal detection device for the wiring relationship of the power metering device further includes a visual feature construction module, a semantic feature construction module, and an undirected graph construction module, where:

[0158] The visual feature construction module is configured to obtain the on-site image data in the area to be detected and input the on-site image data into a pre-constructed visual feature extraction model to obtain the visual features corresponding to the power metering device;

[0159] The semantic feature construction module is configured to obtain the technical parameter description text and physical attribute data of the power metering device, and input the technical parameter description text into a pre-constructed semantic feature extraction module to obtain the semantic features corresponding to the power metering device; the physical attribute data includes wiring connection type data, spatial distance data, and electrical characteristic data;

[0160] The undirected graph construction module is configured to construct an undirected graph of the wiring relationship corresponding to the power metering device according to the visual features, semantic features, and physical attribute data.

[0161] In one embodiment, the undirected graph construction module is further configured to splice the visual features and semantic features to obtain original node features; perform feature dimensionality reduction on the original node features to obtain node features; perform edge feature encoding on the physical attribute data to obtain edge features; and construct an undirected graph of the wiring relationship of the power metering device based on the node features and edge features.

[0162] In one embodiment, the abnormal detection device for the wiring relationship of the electric energy metering device further includes an abnormal detection model training module, which is used to obtain the sample wiring relationship undirected graph of the sample electric energy metering device wiring relationship in the sample area, and obtain the sample fusion features of the sample electric energy metering device wiring relationship according to the sample wiring relationship undirected graph; respectively construct a training data set, a verification data set and a test data set based on the sample fusion features; use the training data set to adjust the model parameters of the abnormal detection model to be trained to obtain a candidate abnormal detection model; adjust the hyperparameters of the candidate abnormal detection model according to the verification data set to obtain the abnormal detection model to be tested; use the test data set to perform performance testing on the abnormal detection model to be tested. If the test result of the performance testing indicates that the abnormal detection model to be tested passes the performance testing, the abnormal detection model to be tested is determined as the abnormal detection model.

[0163] In one embodiment, the abnormal detection model training module is further used to stop parameter adjustment and use the current abnormal detection model to be tested as the abnormal detection model if the test result of the performance testing indicates that the abnormal detection model to be tested fails the performance testing and the number of consecutive failures reaches the set tolerance threshold.

[0164] Each module in the above abnormal detection device for the wiring relationship of the electric energy metering device can be implemented in whole or in part by software, hardware and their combination. The above modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above respective modules.

[0165] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 5 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the wiring relationship undirected graph, fusion features and abnormal detection result data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements an abnormal detection method for the wiring relationship of the electric energy metering device.

[0166] Those skilled in the art can understand that Figure 5 the structure shown in Figure 5 is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0167] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the abnormal detection method for the wiring relationship of the power metering device in the above embodiment is implemented.

[0168] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the abnormal detection method for the wiring relationship of the power metering device in the above embodiment is implemented.

[0169] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the abnormal detection method for the wiring relationship of the power metering device in the above embodiment is implemented.

[0170] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0171] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0172] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0173] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for detecting abnormality in wiring relationship of electric energy metering equipment, characterized in that: The method comprises: Obtaining an undirected wiring relationship graph of the wiring relationship of the electric energy metering equipment in the area to be detected; the undirected wiring relationship graph includes node features and edge features, the node features are features used to characterize the electric energy metering equipment, and the edge features are features used to characterize the connection relationship between the electric energy metering equipment and the circuit element; Inputting the node features and the edge features into a pre-built feature processing model, converting the node features and the edge features into features of the same semantic dimension, and fusing the converted node features and edge features to obtain fused features; The fusion feature is input into an anomaly detection model, and an anomaly detection result of the wiring relationship of the electric energy metering equipment in the area to be detected is obtained through the anomaly detection model; the anomaly detection model is obtained by training the sample fusion feature of the wiring relationship of the sample electric energy metering equipment and the early stopping strategy.

2. The method according to claim 1, characterized in that The feature processing model includes a hierarchical feature extraction module, and the hierarchical feature extraction module includes a first hierarchical feature extraction unit, a second hierarchical feature extraction unit and a third hierarchical feature extraction unit; The step of inputting the node features and the edge features into a pre-built feature processing model, converting the node features and the edge features into features of the same semantic dimension, and fusing the converted node features and edge features to obtain fused features includes: Inputting the node features and edge features into the first hierarchical feature extraction unit to obtain node features and edge features after one extraction; Inputting the node features and edge features extracted once into the second hierarchical feature extraction unit to obtain the node features and edge features extracted twice; Inputting the node features and edge features after the secondary extraction into the third hierarchical feature extraction unit to obtain the node features and edge features after the tertiary extraction; the feature extraction dimensions corresponding to the first hierarchical feature extraction unit, the second hierarchical feature extraction unit and the third hierarchical feature extraction unit decrease in sequence; The node features and edge features extracted three times are converted into features of the same semantic dimension, and the converted node features and edge features extracted three times are fused to obtain the fused features.

3. The method according to claim 2, characterized in that The feature processing model also includes a feature fusion module; The step of converting the node features and edge features extracted three times into features of the same semantic dimension, and fusing the converted node features and edge features extracted three times to obtain the fused features includes: The node features and edge features extracted three times are input into the feature fusion module, and the node features and edge features extracted three times are linearly mapped to obtain feature vectors of the same semantic dimension, and each feature vector is input into the corresponding attention head and linearly mapped to obtain the fused feature.

4. The method according to claim 1, characterized in that Before obtaining the wiring relationship undirected graph of the wiring relationship of the electric energy metering equipment in the area to be detected, the method includes: Acquire on-site image data in the area to be detected, and input the on-site image data into a pre-built visual feature extraction model to obtain visual features corresponding to the electric energy metering device; Obtaining a technical parameter description text and physical property data of the electric energy metering device, and inputting the technical parameter description text into a pre-built semantic feature extraction module to obtain semantic features corresponding to the electric energy metering device; the physical property data includes wiring connection type data, spatial distance data, and electrical characteristic data; A wiring relationship undirected graph corresponding to the wiring relationship of the electric energy metering device is constructed according to the visual features, the semantic features and the physical attribute data.

5. The method according to claim 4, characterized in that The step of constructing a wiring relationship undirected graph corresponding to the wiring relationship of the electric energy metering device according to the visual features, the semantic features and the physical attribute data includes: Concatenating the visual features and the semantic features to obtain original node features; Performing feature dimension reduction on the original node features to obtain the node features; Performing edge feature encoding on the physical attribute data to obtain the edge feature; Based on the node features and the edge features, an undirected wiring relationship graph of the wiring relationship of the electric energy metering equipment is constructed.

6. The method according to claim 1, characterized in that The anomaly detection model is trained by the following steps: Obtaining a sample wiring relationship undirected graph of wiring relationships of sample electric energy metering devices in a sample area, and obtaining sample fusion features of the wiring relationships of the sample electric energy metering devices according to the sample wiring relationship undirected graph; Based on the sample fusion features, a training data set, a validation data set and a test data set are respectively constructed; Using the training data set to adjust the model parameters of the anomaly detection model to be trained, to obtain a candidate anomaly detection model; Adjusting the hyperparameters of the candidate anomaly detection model according to the verification data set to obtain an anomaly detection model to be tested; The test data set is used to perform a performance test on the anomaly detection model to be tested. If a test result of the performance test indicates that the anomaly detection model to be tested passes the performance test, the anomaly detection model to be tested is determined as the anomaly detection model.

7. The method according to claim 6, characterized in that The method further comprises: If the test result of the performance test indicates that the anomaly detection model to be tested has not passed the performance test, and the number of consecutive failures reaches the set tolerance threshold, parameter adjustment is stopped and the current anomaly detection model to be tested is used as the anomaly detection model.

8. A device for detecting abnormality in wiring relationship of electric energy metering equipment, characterized in that: The device comprises: An undirected graph acquisition module, used to acquire a wiring relationship undirected graph of the wiring relationship of the electric energy metering equipment in the area to be detected; the wiring relationship undirected graph contains node features and edge features, the node features are features used to characterize the electric energy metering equipment, and the edge features are features used to characterize the connection relationship between the electric energy metering equipment and the circuit elements; A feature acquisition module, used for inputting the node features and the edge features into a pre-built feature processing model, converting the node features and the edge features into features of the same semantic dimension, and fusing the converted node features and edge features to obtain fused features; The anomaly detection module is used to input the fusion feature into an anomaly detection model, and obtain the anomaly detection result of the wiring relationship of the electric energy metering equipment in the area to be detected through the anomaly detection model; the anomaly detection model is obtained by training the sample fusion feature of the wiring relationship of the sample electric energy metering equipment and the early stopping strategy.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images