Electric energy metering equipment wiring rectification suggestion generation method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to the technical field of automatic operation and maintenance and intelligent decision making of electric energy metering equipment, and provides a method and device for generating a wiring rectification suggestion of the electric energy metering equipment, computer equipment, a readable storage medium and a program product. The method comprises the following steps: performing feature extraction on historical wiring fault text data and historical wiring fault image data of the electric energy metering equipment to obtain historical wiring fault text features and historical wiring fault image features; performing feature extraction on the historical wiring fault time sequence data after the position information is injected to obtain historical wiring fault time sequence features; performing feature fusion on the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time sequence features to obtain cross-modal fusion features; and obtaining a wiring rectification suggestion of the electric energy metering equipment according to the cross-modal fusion features and a suggestion generation model. By adopting the method, the globality and systematicness of the wiring rectification suggestion of the electric energy metering equipment can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of automatic operation and maintenance and intelligent decision-making of power metering equipment, and particularly to a method, device, computer device, computer-readable storage medium, and computer program product for generating wiring rectification suggestions for power metering equipment. Background Art

[0002] In recent years, with the transformation of the power system towards digitalization, intelligentization, and networking, power metering equipment, as an important supporting facility for grid informatization, the wiring status of power metering equipment is directly related to the accuracy of power metering and power supply safety. However, due to the complex on-site environment, numerous historical operation and maintenance records, and the dependence of traditional inspection methods on manual judgment, some wiring installations have defects or errors, resulting in low operation and maintenance efficiency and an extended fault troubleshooting cycle. Therefore, it is of great significance to rectify the wiring of power metering equipment.

[0003] Traditional methods for generating wiring rectification suggestions for power metering equipment perform joint modeling on image data and text data to obtain wiring rectification suggestions. However, traditional methods for generating wiring rectification suggestions for power metering equipment are limited to recurrent neural networks (RNNs) or long short-term memory networks (LSTMs) in terms of temporal information modeling, suffering from insufficient long-distance dependence capture capabilities, resulting in a lack of globality and systematicness in the wiring rectification suggestions for power metering equipment. Summary of the Invention

[0004] Based on this, it is necessary to provide a method, device, computer device, computer-readable storage medium, and computer program product for generating wiring rectification suggestions for power metering equipment to address the above technical problems.

[0005] In a first aspect, the present application provides a method for generating wiring rectification suggestions for power metering equipment, including:

[0006] Obtaining historical wiring fault text data, historical wiring fault image data, and historical wiring fault temporal data of the power metering equipment;

[0007] Performing feature extraction on the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features;

[0008] Injecting position information into the historical wiring fault temporal data, and extracting correlation information of the historical wiring fault temporal data after injecting the position information according to the self-attention mechanism in the transformer encoder to obtain historical wiring fault temporal features;

[0009] Perform feature fusion on the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain cross-modal fusion features;

[0010] According to the cross-modal fusion features and the recommendation generation model, obtain the wiring rectification recommendations for the power metering device.

[0011] In one embodiment, the feature extraction of the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features includes:

[0012] According to the bidirectional encoder representation model based on the transducer, perform feature extraction on the historical wiring fault text data to obtain historical wiring fault text features;

[0013] According to the dense connection network model, perform feature extraction on the historical wiring fault image data to obtain historical wiring fault image features.

[0014] In one embodiment, the feature fusion of the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain cross-modal fusion features includes:

[0015] Calculate the respective weights of the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features according to the gating mechanism;

[0016] According to the respective weights of the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features, perform weighted fusion on the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain cross-modal fusion features.

[0017] In one embodiment, the obtaining of the wiring rectification recommendations for the power metering device according to the cross-modal fusion features and the recommendation generation model includes:

[0018] Input the cross-modal fusion features into the recommendation generation model to obtain candidate wiring rectification recommendations;

[0019] According to the pre-constructed power terminology dictionary, check whether each candidate word in the candidate wiring rectification recommendations belongs to the power terminology dictionary;

[0020] When there is a candidate word that does not belong to the power terminology dictionary, remove the candidate word from the candidate wiring rectification recommendations, or replace the candidate word with a word with similar semantics in the power terminology dictionary to obtain the wiring rectification recommendations for the power metering device.

[0021] In one embodiment, after obtaining the wiring rectification suggestion of the power metering device, the method further includes:

[0022] Obtaining a feedback vector according to user feedback in the actual operation and maintenance process of the wiring rectification of the power metering device;

[0023] Inputting the feedback vector into a gated recurrent unit model to obtain a hidden state vector;

[0024] Optimizing the suggestion generation model according to the hidden state vector.

[0025] In one embodiment, the historical wiring fault text data of the power metering device includes alarm logs, manual inspection records, and work order system data; the historical wiring fault image data of the power metering device includes on-site photos of wiring errors; the historical wiring fault time series data of the power metering device includes historical error occurrence frequencies and device status time series indicators.

[0026] In a second aspect, the present application further provides a device for generating wiring rectification suggestions for a power metering device, including:

[0027] A wiring fault data acquisition module, configured to acquire historical wiring fault text data, historical wiring fault image data, and historical wiring fault time series data of the power metering device;

[0028] A wiring fault feature extraction module, configured to extract features from the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features;

[0029] The wiring fault feature extraction module is further configured to inject position information into the historical wiring fault time series data, and extract correlation information of the historical wiring fault time series data after injecting the position information according to the self-attention mechanism in the transformer encoder to obtain historical wiring fault time series features;

[0030] A feature fusion module, configured to perform feature fusion on the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain cross-modal fusion features;

[0031] A wiring rectification suggestion generation module, configured to obtain the wiring rectification suggestion of the power metering device according to the cross-modal fusion features and the suggestion generation model.

[0032] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor executes the above method.

[0033] Fourthly, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the above method.

[0034] Fifthly, the present application also provides a computer program product. The computer program product includes a computer program, and the computer program is executed by a processor to implement the above method.

[0035] For the above method, device, computer equipment, computer-readable storage medium, and computer program product for generating wiring rectification suggestions for power metering equipment, historical wiring fault text data, historical wiring fault image data, and historical wiring fault time series data of the power metering equipment are obtained; feature extraction is performed on the historical wiring fault text data and historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features; position information is injected into the historical wiring fault time series data, and according to the self-attention mechanism in the transformer encoder, the correlation information of the historical wiring fault time series data after injecting the position information is extracted to obtain historical wiring fault time series features; feature fusion is performed on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features to obtain cross-modal fusion features; and wiring rectification suggestions for the power metering equipment are obtained according to the cross-modal fusion features and the suggestion generation model. In the present application, position information is injected into the historical wiring fault time series data, and according to the self-attention mechanism in the transformer encoder, the correlation information of the historical wiring fault time series data after injecting the position information is extracted to obtain historical wiring fault time series features, which can effectively capture the long-range correlation in the historical wiring fault time series data, thereby improving the globality and systematicness of the wiring rectification suggestions for the power metering equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] 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, other related drawings can be obtained without creative efforts based on these drawings.

[0037] Figure 1 It is an application environment diagram of the method for generating wiring rectification suggestions for power metering equipment in an embodiment;

[0038] Figure 2 It is a flowchart of the method for generating wiring rectification suggestions for power metering equipment in an embodiment;

[0039] Figure 3 It is a flowchart of the method for generating wiring rectification suggestions for power metering equipment in another embodiment;

[0040] Figure 4 It is a structural block diagram of a device for generating wiring rectification suggestions for an electric energy metering device in an embodiment;

[0041] Figure 5 It is an internal structure diagram of a computer device in an embodiment. Specific implementation manners

[0042] In order to make the objectives, technical solutions and advantages of the present application clearer and more 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.

[0043] The method for generating wiring rectification suggestions for an electric energy metering device provided by an embodiment of the present application can be applied to an application environment as shown in Figure 1 wherein, the terminal 102 communicates with the server 104 through a network. 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 104, or can be placed in the cloud or other network servers. The terminal 102 can obtain the historical wiring fault text data, historical wiring fault image data and historical wiring fault time series data of the electric energy metering device, and then obtain the wiring rectification suggestions for the electric energy metering device. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The server 104 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.

[0044] In an exemplary embodiment, as shown in Figure 2 a method for generating wiring rectification suggestions for an electric energy metering device is provided. Taking the method applied to the terminal 102 in Figure 1 as an example, the method includes the following steps S201 to S205.

[0045] Step S201, obtain the historical wiring fault text data, historical wiring fault image data and historical wiring fault time series data of the electric energy metering device.

[0046] The alarm logs, manual inspection records and work order system data at the time of wiring faults of the electric energy metering device can be collected as the historical wiring fault text data of the electric energy metering device; among them, the alarm logs include time stamps, device IDs and error type information, and the manual inspection records can be described in natural language; the work order system data includes the historical wiring fault phenomena of the electric energy metering device and the processing processes corresponding to each historical wiring fault phenomenon.

[0047] On-site photos of wiring errors when the wiring of the electric energy metering device fails can be collected as historical wiring fault image data of the electric energy metering device; wherein the on-site photos of the wiring errors use a red-green-blue color model (RGB) format with a resolution greater than or equal to 1024×768.

[0048] The historical wiring fault timing data of the electric energy metering device includes the ability to obtain historical error occurrence frequencies and device status timing indicators; wherein the device status timing indicators may include voltage indicators, current indicators, and temperature indicators.

[0049] Step S202 , extracting features from the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features.

[0050] After obtaining the historical wiring fault text data, historical wiring fault image data and historical wiring fault time series data of the electric energy metering device, the historical wiring fault text data and the historical wiring fault time series data may be text cleansed, and the historical wiring fault image data may be standardized.

[0051] The steps for text cleaning of historical wiring fault text data and historical wiring fault time series data are as follows:

[0052] Regular expressions are used to remove non-Chinese characters from the historical wiring fault text data and the historical wiring fault time series data. Low-frequency noise words in the historical wiring fault text data and the historical wiring fault time series data are eliminated based on the term frequency-inverse document frequency (TF-IDF), and the words with term frequency-inverse document frequency values ​​higher than the threshold τ = 0.01 are retained, as shown in formula (1).

[0053] (1)

[0054] in, Represents the words in the historical wiring fault text data, TF ( ) indicates a word The frequency of occurrence in the historical wiring fault text data, DF ( ) indicates a word The rarity in the historical wiring fault text data, N represents the total number of documents in the historical wiring fault text data.

[0055] The steps for standardizing the historical wiring fault image data are as follows:

[0056] The historical wiring fault image data is histogram equalized and normalized, as shown in formula (2).

[0057] (2)

[0058] Among them, Inorm is the pixel value after normalizing the historical wiring fault image, I is the pixel value of the historical wiring fault image, is the pixel mean value of the historical wiring fault image, is the pixel variance value of the historical wiring fault image.

[0059] It is possible to extract highly discriminative features from the historical wiring fault text data after text cleaning and the historical wiring fault image data after normalization processing, obtaining historical wiring fault text features and historical wiring fault image features.

[0060] Step S203: Inject position information into the historical wiring fault time-series data, and according to the self-attention mechanism in the transformer encoder, extract the correlation information of the historical wiring fault time-series data after injecting the position information, obtaining historical wiring fault time-series features.

[0061] It is possible to inject position information into the historical wiring fault time-series data, as shown in Equation (3).

[0062] (3)

[0063] Among them, PE represents the trainable position embedding matrix, pos represents the position, i represents the dimension index, d represents the dimension, 10000 is a hyperparameter used to adjust the wavelength range, and 10000 2i / d represents the wavelength change.

[0064] It is possible to extract the correlation information of the historical wiring fault time-series data after injecting the position information according to the self-attention mechanism in the transformer encoder, obtaining historical wiring fault time-series features , = 512.

[0065] Among them, the multi-head attention calculation formula is as shown in Equation (4) (taking head k as an example).

[0066] (4)

[0067] Among them, Q is the query vector, K is the key vector, V is the value vector, and dk is the scaling factor.

[0068] Step S204: Perform feature fusion on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time-series features to obtain cross-modal fusion features.

[0069] After obtaining the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features, the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features can be feature-aligned to construct a spatio-temporal alignment matrix , where N is the number of power metering devices, T is the time window, and the element represents the multi-modal feature set of power metering device i at time slice j, where the multi-modal features include historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features.

[0070] After feature-aligning the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features, according to the importance of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features respectively, the gating mechanism can be used to dynamically assign weights to the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features, and according to the respective weights of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features, the historical wiring fault text features, historical wiring fault image features, and historical wiring fault time series features can be weighted and aggregated to achieve adaptive cross-modal feature fusion and obtain cross-modal fusion features.

[0071] Step S205, according to the cross-modal fusion features and the recommendation generation model, obtain the wiring rectification recommendations for the power metering device.

[0072] The recommendation generation model can be obtained according to the Transformer Decoder.

[0073] The cross-modal fusion features can be input into the recommendation generation model to obtain the input result of the recommendation generation model; the wiring rectification recommendations for the power metering device can be obtained according to the input result of the above recommendation generation model.

[0074] In the above method for generating the wiring rectification recommendations for the power metering device, position information is injected into the historical wiring fault time series data, and according to the self-attention mechanism in the Transformer encoder, the correlation information of the historical wiring fault time series data after injecting the position information is extracted to obtain the historical wiring fault time series features, which can effectively capture the long-range correlation in the historical wiring fault time series data, thereby improving the globality and systematicness of the wiring rectification recommendations for the power metering device.

[0075] In one embodiment, feature extraction is performed on historical wiring fault text data and historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features. The specific steps are as follows: According to the Bidirectional Encoder Representations from Transformers (BERT) model, feature extraction is performed on the historical wiring fault text data to obtain historical wiring fault text features; according to the Densely Connected Convolutional Networks (DenseNet) model, feature extraction is performed on the historical wiring fault image data to obtain historical wiring fault image features.

[0076] The Bidirectional Encoder Representations from Transformers (BERT) can be the BERT-base model.

[0077] Feature extraction can be performed on the historical wiring fault text data according to the Bidirectional Encoder Representations from Transformers (BERT) model to obtain historical wiring fault text features. The specific steps are as follows:

[0078] For the input text Generate context encoding as shown in Equation (5).

[0079] (5)

[0080] where T represents the input text, and E text represents the context encoding of the input text.

[0081] The [CLS] token vector can be used as the global semantic representation to obtain the historical wiring fault text features .

[0082] The Densely Connected Convolutional Networks (DenseNet) can be the DenseNet-121 model.

[0083] Feature extraction can be performed on the historical wiring fault image data according to the DenseNet-121 model, and the historical wiring fault image features can be obtained through global average pooling . Among them, the output result of the th layer of the DenseNet-121 model is as shown in Equation (6).

[0084] (6)

[0085] where Represents a triple of Batch Normalization - Rectified Linear Unit – Convolution (BN - ReLU - Conv).

[0086] In this embodiment, based on the Transformer - based Bidirectional Encoder Representations from Transformers (BERT) model, feature extraction is performed on the historical wiring fault text data to obtain relatively accurate historical wiring fault text features; based on the Dense Connectivity Network (DenseNet) model, feature extraction is performed on the historical wiring fault image data to obtain relatively accurate historical wiring fault image features, providing rich feature information for subsequent generation of wiring rectification suggestions.

[0087] In one embodiment, feature fusion is performed on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features to obtain cross - modal fusion features. The specific steps are as follows: Calculate the respective weights of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features according to the gating mechanism; perform weighted fusion on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features according to their respective weights to obtain cross - modal fusion features.

[0088] The importance of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features can be used to learn weight vectors for the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features using the gating mechanism, dynamically assign weights to the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features, and obtain the respective weights of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features, as shown in Equation (7).

[0089] (7)

[0090] Where represents the weight,[[]] represents the Sigmoid function, W g represents the weight matrix,[[]] , b g represents the bias term.

[0091] Perform weighted fusion on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features according to their respective weights to obtain cross - modal fusion features, as shown in Equation (8).

[0092] (8)

[0093] Among them, represents the cross-modal fusion feature, represents the historical wiring fault image feature, represents the weight corresponding to the historical wiring fault image feature, represents the historical wiring fault text feature, represents the weight corresponding to the historical wiring fault text feature, represents the historical wiring fault time series feature, represents the weight corresponding to the historical wiring fault time series feature.

[0094] In this embodiment, according to the gating mechanism, the weights corresponding to the historical wiring fault text feature, the historical wiring fault image feature, and the historical wiring fault time series feature are calculated to perform weighted fusion on the historical wiring fault text feature, the historical wiring fault image feature, and the historical wiring fault time series feature, so as to obtain the cross-modal fusion feature. It can deeply explore the correlation between image information and text information, make full use of the complementary information between different modalities, improve the device state understanding ability of the recommendation generation model for power metering devices and the accuracy of wiring fault diagnosis, and thus improve the accuracy of the wiring rectification suggestions for power metering devices.

[0095] In one of the embodiments, according to the cross-modal fusion feature and the recommendation generation model, the wiring rectification suggestion for the power metering device is obtained. The specific steps are as follows: input the cross-modal fusion feature into the recommendation generation model to obtain candidate wiring rectification suggestions; according to the pre-constructed power terminology dictionary, check whether each candidate word in the candidate wiring rectification suggestions belongs to the power terminology dictionary; when there is a candidate word that does not belong to the power terminology dictionary, remove the candidate word from the candidate wiring rectification suggestions, or replace the candidate word with a word with similar semantics in the power terminology dictionary to obtain the wiring rectification suggestion for the power metering device.

[0096] The Transformer Decoder can be used as the recommendation generation model.

[0097] Among them, the probability that the Transformer Decoder generates the word at time step t is shown in Equation (9).

[0098] (9)

[0099] Among them, represents the probability that the Transformer Decoder generates the word at time step t, represents the cross-modal fusion feature, and softmax represents the normalized exponential function, denotes the weight matrix of the output layer, and is the hidden state of the decoder at the L-th layer.

[0100] The cross-modal fusion features can be input into the recommendation generation model to obtain candidate wiring rectification recommendations.

[0101] After obtaining the candidate wiring rectification recommendations, a constraint generation mechanism can be introduced. According to the pre-constructed power terminology dictionary, check whether each candidate word in the candidate wiring rectification recommendations belongs to the power terminology dictionary; when there is a candidate word that does not belong to the power terminology dictionary, remove the candidate word from the candidate wiring rectification recommendations, or replace the candidate word with a word with similar semantics in the power terminology dictionary to obtain the wiring rectification recommendations for the power metering equipment.

[0102] In this embodiment, the cross-modal fusion features are input into the recommendation generation model to obtain candidate wiring rectification recommendations; when there is a candidate word in the candidate wiring rectification recommendations that does not belong to the pre-constructed power terminology dictionary, remove the candidate word from the candidate wiring rectification recommendations, or replace the candidate word with a word with similar semantics in the power terminology dictionary to obtain the wiring rectification recommendations for the power metering equipment, thereby ensuring the professionalism and practicality of the generated wiring rectification recommendations.

[0103] In one of the embodiments, after obtaining the wiring rectification recommendations for the power metering equipment, the method provided by this application further includes: obtaining a feedback vector according to the user feedback in the actual operation and maintenance process of the wiring rectification of the power metering equipment; inputting the feedback vector into a gated recurrent unit model to obtain a hidden state vector; and optimizing the recommendation generation model according to the hidden state vector.

[0104] The user feedback in the actual operation and maintenance process of the wiring rectification of the power metering equipment can be collected to construct a closed-loop feedback mechanism to continuously optimize the recommendation generation model. Specifically, the user feedback in the actual operation and maintenance process of the wiring rectification of the power metering equipment can be automatically or semi-automatically collected from the operation records of the maintenance personnel, the user evaluation interface, and the log files.

[0105] The user feedback in the actual operation and maintenance process of the wiring rectification of the power metering equipment can be quantified into a numerical form for the model to learn to obtain a feedback vector , thereby effectively utilizing the user feedback information. The dimension m of the feedback vector can be extended according to the requirements of the actual application scenario to include more types of feedback information.

[0106] The feedback vector can be input into the gated recurrent unit model to obtain the hidden state vector. Among them, the Gated Recurrent Unit (GRU) is a commonly used structure of the Recurrent Neural Network (RNN). By introducing a gating mechanism to control the flow of information, it can effectively capture the long-term dependencies in sequential data.

[0107] Based on the hidden state vector, the parameters of the recommendation generation model can be optimized, such as the learning rate and regularization coefficient. The specific formula for iteratively updating the model parameters is shown in Equation (10).

[0108] (10)

[0109] Among them, z t represents the update gate, r t represents the input at the current moment, represents the hidden state at the previous moment, h t represents the hidden state at the current moment, Wz, W h represents the weight matrix, represents the Sigmoid function, and tanh represents the hyperbolic tangent function.

[0110] In this embodiment, according to the user feedback in the actual operation and maintenance process of the wiring rectification of the electric energy metering device, a feedback vector is obtained; the feedback vector is input into the gated recurrent unit model to obtain the hidden state vector; and based on the hidden state vector, the recommendation generation model is optimized. It can achieve the continuous iterative update of the recommendation generation model and the continuous improvement of performance, thereby realizing the dynamic optimization and intelligent evolution of the wiring rectification recommendations.

[0111] To better understand the above method, the following details an application embodiment of the method for generating wiring rectification recommendations for the electric energy metering device of this application, as Figure 3 shown.

[0112] The technical solution provided by this embodiment relates to the field of automatic operation and maintenance and intelligent decision-making of power metering equipment, especially focusing on the intelligent generation technology of wiring rectification suggestions. In recent years, with the transformation of the power system towards digitalization, intelligence, and networking, as an important supporting facility for grid informatization, the wiring status of power metering equipment is directly related to the accuracy of power metering and power supply safety. However, due to the complex on-site environment, numerous historical operation and maintenance records, and the dependence on manual judgment in traditional inspection methods, some wiring installations have defects or errors, leading to problems such as low operation and maintenance efficiency and extended fault troubleshooting cycles. In response to this situation, the industry's demand for comprehensive intelligent analysis technologies that can integrate image information, text records, and historical error data is increasing. At the same time, with the continuous breakthroughs in deep learning technology and natural language processing technology, especially the superior characteristics shown by the self-attention mechanism and multi-modal convolutional neural network in complex data scenarios, it provides a new technical idea and implementation path for solving the wiring rectification problem of power metering equipment.

[0113] The current intelligent detection technologies for the wiring of power metering equipment are mainly divided into two categories: one is the single-modal analysis method based on image recognition, which extracts the wiring image features through Convolutional Neural Networks (CNN) to achieve error detection. For example, the improved Residual Network (ResNet) or YOLO (You Only Look Once) algorithm is used to identify the status of wiring terminals, but it only focuses on static image features and lacks the correlation analysis of temporal information such as historical operation and maintenance records and text descriptions. The other is the rule inference system based on knowledge graphs, which generates suggestions by constructing an expert knowledge base to match preset rules. However, such systems rely on manually labeled logical rules, are difficult to adapt to complex and changeable on-site scenarios, and cannot achieve dynamic optimization.

[0114] With the development of multi-modal learning technology, some studies have attempted to jointly model image and text data. For example, in industrial equipment fault diagnosis, a two-stream network is used to fuse sensor data and maintenance logs, or in medical image analysis, CT (Computed Tomography) images are combined with medical record texts to generate diagnostic reports. However, the traditional methods for generating wiring rectification suggestions for power metering equipment mostly adopt simple splicing or average pooling strategies in the feature fusion stage, failing to fully explore the deep semantic associations of cross-modal features, and still being limited to the Recurrent Neural Network (RNN) or Long Short-Term Memory (LSTM) in the temporal information modeling, with the problem of insufficient ability to capture long-distance dependencies.

[0115] In recent years, the converter architecture has shown significant advantages in time series modeling tasks due to its global self-attention mechanism, especially in the field of power equipment state prediction. Studies have used converter encoders to process time series data of equipment operating parameters, effectively improving the accuracy of anomaly detection. At the same time, pre-trained language models represented by the GPT series have promoted breakthroughs in natural language generation technology, making end-to-end generation from structured data to text descriptions possible. However, how to deeply integrate the above technologies with multimodal features to build a closed-loop optimization system with dynamic perception capabilities is still a technical problem that needs to be solved.

[0116] The innovation of the technical solution provided by this embodiment lies in: combining the self-attention mechanism with the multimodal convolutional neural network, performing time series modeling on the historical wiring error information through the transformer encoder, obtaining the time series features of the historical wiring fault, using the dense connection network and the transformer-based bidirectional encoder representation model to extract the high-dimensional features of the historical wiring fault image data and the historical wiring fault text data respectively, and designing a weighted fusion strategy to realize the dynamic interaction of cross-modal features, and obtain cross-modal fusion features. On this basis, personalized rectification suggestions are output through the transformer-based natural language generation model, and an online feedback mechanism is introduced to form a closed-loop system for continuous optimization. Compared with traditional technologies, the technical solution provided by this embodiment not only breaks through the limitations of single modal analysis, but also realizes the collaborative reasoning of historical experience and real-time status through the self-attention mechanism, providing the power industry with a high-precision and highly adaptable intelligent operation and maintenance solution, which meets the core demands of intelligent and refined equipment management in the era of energy Internet.

[0117] Traditional methods for generating rectification suggestions for wiring of electric energy metering equipment face multiple technical bottlenecks. Insufficient capture of long-distance dependencies results in a lack of globality in rectification suggestions, which cannot fully consider historical data and long-term trends; the degree of multimodal data fusion is low, making it difficult to deeply explore the relationship between image and text information, which seriously affects the accuracy of rectification suggestions; there is a general lack of feedback and self-optimization mechanisms, and intelligent evolution is slow; and traditional methods lack real-time performance and processing efficiency when processing massive amounts of data. These problems restrict the level of intelligent operation and maintenance and processing efficiency. To solve the above problems, it is necessary to introduce more advanced deep learning models, improve the depth of multimodal fusion, establish a feedback optimization mechanism, and realize real-time, accurate and intelligent generation of rectification suggestions.

[0118] The technical solution provided in this embodiment innovatively integrates the self-attention mechanism and the recurrent neural network. Aiming at the deficiencies of the traditional method for generating wiring rectification suggestions for power metering equipment in capturing long-distance dependencies, multi-modal data fusion, and intelligent evolution, the main innovation points of the technical solution provided in this embodiment are as follows: A transformer self-attention mechanism-based encoder is proposed to effectively capture the long-range correlations in historical operation and maintenance data, enhancing the globality and systematicness of rectification suggestions; A multi-modal fusion module is designed to deeply fuse wiring images and text description information, significantly improving the accuracy of rectification suggestions; And a feedback loop mechanism based on the recurrent neural network is introduced to achieve the dynamic optimization and intelligent evolution of rectification suggestions, enabling the model to continuously learn and adapt to new operation and maintenance scenarios, and finally realizing the generation of more efficient, accurate, and intelligent wiring rectification suggestions for power metering equipment, improving the level of operation and maintenance intelligence and the overall processing efficiency.

[0119] The method for generating wiring rectification suggestions for power metering equipment provided in this embodiment includes the following steps:

[0120] Step S1: Multi-modal historical operation and maintenance data collection and preprocessing.

[0121] The goal of this step is to standardize multi-source heterogeneous data and construct high-quality multi-modal historical operation and maintenance data.

[0122] 1.1 Data collection

[0123] Alarm logs, manual inspection records, and work order system data during wiring faults of power metering equipment can be collected as historical wiring fault text data of power metering equipment; among them, the alarm logs include timestamp, device ID, and error type information, and the manual inspection records can be described in natural language; the work order system data includes historical wiring fault phenomena of power metering equipment and the corresponding processing processes for each historical wiring fault phenomenon.

[0124] Photos of the wiring error site during wiring faults of power metering equipment can be collected as historical wiring fault image data of power metering equipment; among them, the photos of the wiring error site are in the Red Green Blue (RGB) color model format with a resolution greater than or equal to 1024×768.

[0125] The historical wiring fault time series data of power metering equipment includes the historical error occurrence frequency and device status time series indicators that can be obtained; among them, the device status time series indicators can include voltage indicators, current indicators, and temperature indicators.

[0126] 1.2 Data cleaning

[0127] After obtaining the historical wiring fault text data, historical wiring fault image data, and historical wiring fault time series data of the power metering device, text cleaning can be performed on the historical wiring fault text data and historical wiring fault time series data, and standardization processing can be performed on the historical wiring fault image data.

[0128] The steps for text cleaning the historical wiring fault text data and historical wiring fault time series data are as follows:

[0129] Use regular expressions to remove non-Chinese characters from the historical wiring fault text data and historical wiring fault time series data, and based on Term Frequency-Inverse Document Frequency (TF-IDF), remove low-frequency noise words from the historical wiring fault text data and historical wiring fault time series data, retaining words with a TF-IDF value higher than the threshold τ = 0.01, as shown in Equation (1).

[0130] (1)

[0131] Where, represents the word in the historical wiring fault text data, TF( ) represents the word appearance frequency in the historical wiring fault text data, DF( ) represents the word rarity in the historical wiring fault text data, and N represents the total number of documents in the historical wiring fault text data.

[0132] The steps for standardizing the historical wiring fault image data are as follows:

[0133] Perform histogram equalization and normalization on the historical wiring fault image data, as shown in Equation (2).

[0134] (2)

[0135] Where, Inorm is the pixel value after standardizing the historical wiring fault image, I is the pixel value of the historical wiring fault image, is the pixel mean of the historical wiring fault image, is the pixel variance value of the historical wiring fault image.

[0136] Step S2: Multimodal feature extraction.

[0137] This step aims to extract highly discriminative features from the historical wiring fault text data and historical wiring fault image data of the power metering device. It includes two parts: image feature extraction and text feature extraction.

[0138] Based on the Bidirectional Encoder Representations from Transformers (BERT), the historical wiring fault text data can be used for feature extraction to obtain historical wiring fault text features. The specific steps are as follows:

[0139] For the input text Generate context encoding as shown in Equation (5).

[0140] (5)

[0141] where T represents the input text, and E text represents the context encoding of the input text.

[0142] The [CLS] token vector can be used as the global semantic representation to obtain the historical wiring fault text features .

[0143] Among them, the Bidirectional Encoder Representations from Transformers (BERT) can be the BERT-base model.

[0144] Based on the DenseNet-121 model, the historical wiring fault image data can be used for feature extraction, and the historical wiring fault image features can be obtained through global average pooling . Among them, the output result of the layer of the DenseNet-121 model is as shown in Equation (6).

[0145] (6)

[0146] where represents the Batch Normalization - Rectified Linear Unit – Convolution (BN-ReLU-Conv) triple.

[0147] Among them, the DenseNet model can be the Densely Connected Convolutional Networks-121 (DenseNet-121) model.

[0148] Step S3: Temporal dependence modeling.

[0149] This step aims to capture the long - distance temporal dependencies in the historical wiring fault temporal data, so as to more accurately understand the evolution of the power metering device state and the occurrence law of wiring faults.

[0150] Location information can be injected into the historical wiring fault temporal data, as shown in Equation (3).

[0151] (3)

[0152] Where, PE represents the trainable position embedding matrix, pos represents the position, i represents the dimension index, d represents the dimension, 10000 is a hyperparameter used to adjust the wavelength range, and 10000 2i / d represents the wavelength change.

[0153] According to the self - attention mechanism in the Transformer Encoder, the correlation information of the historical wiring fault temporal data after injecting location information can be extracted to obtain the historical wiring fault temporal features , = 512.

[0154] Where the formula for multi - head attention is shown in Equation (4) (taking head k as an example).

[0155] (4)

[0156] Where, Q is the query vector, K is the key vector, V is the value vector, and dk is the scaling factor.

[0157] Temporal features Capture the long - distance temporal dependencies and provide rich semantic information for subsequent wiring fault diagnosis and rectification suggestion generation.

[0158] Step S4: Cross - modal dynamic fusion.

[0159] This step can achieve the effective fusion of image - text - temporal features, make full use of the complementary information between different modalities, and improve the understanding ability of the suggestion generation model for the power metering device state and the accuracy of wiring fault diagnosis. The core idea of this step is to dynamically assign weights according to the importance of different modality features and perform weighted aggregation, so as to achieve adaptive cross - modal feature fusion.

[0160] Traditional feature fusion methods usually adopt simple concatenation or average pooling, ignoring the heterogeneity and importance differences of different modality features. To solve this problem, this step introduces an adaptive weight assignment strategy based on a gating mechanism. The core idea of this strategy is to learn a weight vector for each modality, and this weight vector represents the importance of the modality features at the current moment.

[0161] Specifically, according to the importance of the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features respectively, a gating mechanism can be used to learn weight vectors for the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features, dynamically assign weights to the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features, and obtain the weights corresponding to the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features respectively, as shown in Equation (7).

[0162] (7)

[0163] Among them, represents the weight, represents the Sigmoid function, W g represents the weight matrix, , b g represents the bias term.

[0164] According to the weights corresponding to the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features respectively, perform weighted fusion on the historical wiring fault text features, historical wiring fault image features, and historical wiring fault temporal features to obtain cross-modal fusion features, as shown in Equation (8).

[0165] (8)

[0166] Among them, represents the cross-modal fusion feature, represents the historical wiring fault image feature, represents the weight corresponding to the historical wiring fault image feature, represents the historical wiring fault text feature, represents the weight corresponding to the historical wiring fault text feature, represents the historical wiring fault temporal feature, represents the weight corresponding to the historical wiring fault temporal feature.

[0167] In this step, according to the importance of different modal features, their contributions in the final feature representation can be adaptively adjusted, so as to achieve more effective cross-modal information fusion. This dynamic weighted fusion strategy can better utilize the complementary information between different modalities and improve the understanding ability of the recommendation generation model for complex scenarios.

[0168] Step S5: Generation of rectification suggestions.

[0169] Based on the fused cross-modal fusion features, this step can generate accurate, fluent, and power industry standard-compliant wiring rectification suggestions in text form.

[0170] The Transformer Decoder can be used as the suggestion generation model.

[0171] Among them, the probability of the Transformer Decoder generating a word at time step t is shown in Equation (9).

[0172] (9)

[0173] Among them, represents the probability of the Transformer Decoder generating a word at time step t , represents the cross-modal fusion features, softmax represents the normalized exponential function, represents the weight matrix of the output layer, is the hidden state of the L-th layer decoder.

[0174] The cross-modal fusion features can be input into the suggestion generation model to obtain candidate wiring rectification suggestions.

[0175] After obtaining the candidate wiring rectification suggestions, a constraint generation mechanism can be introduced to filter the candidate wiring rectification suggestions to obtain the wiring rectification suggestions, thereby ensuring that the generated wiring rectification suggestion text complies with the power industry standards and contains accurate professional terms.

[0176] Specifically, during the beam search process, the generated candidate words can be filtered using a pre-constructed power term dictionary. Beam search is a commonly used text generation algorithm that can retain the k most likely candidate sequences at each time step (k is the beam size). In traditional beam search, the selection of candidate words is only based on the prediction probability of the model. In the constraint generation mechanism in this step, it can be further checked whether each candidate word belongs to the power term dictionary. When there is a candidate word that does not belong to the power term dictionary, the candidate word is removed from the candidate wiring rectification suggestions, or the candidate word is replaced with a semantically similar word in the power term dictionary to obtain the wiring rectification suggestions for the electrical energy metering device, thereby ensuring the accuracy of the generated wiring rectification suggestions.

[0177] Step S6: Online feedback optimization.

[0178] It is possible to collect user feedback during the actual operation and maintenance of the wiring rectification of the electric energy metering device, and construct a closed-loop feedback mechanism to continuously optimize the recommendation generation model. The core of this step is to quantify user feedback by constructing a feedback vector, and use a gated recurrent unit to dynamically adjust the parameters of the recommendation generation model to achieve adaptive optimization of the recommendation generation model.

[0179] The user feedback during the actual operation and maintenance of the wiring rectification of the electric energy metering device can be quantified into a numerical form that can be learned by the model to obtain a feedback vector , so as to effectively utilize user feedback information. The dimension m of the feedback vector can be extended according to the requirements of the actual application scenario to include more types of feedback information. Among them, the user feedback during the actual operation and maintenance of the wiring rectification of the electric energy metering device can be collected automatically or semi-automatically from the operation records of maintenance personnel, the user evaluation interface, and the log files.

[0180] The feedback vector can be input into the gated recurrent unit model to obtain a hidden state vector. Among them, the gated recurrent unit (GRU) is a commonly used structure of the recurrent neural network (RNN). By introducing a gating mechanism to control the flow of information, it can effectively capture the long-term dependencies in sequential data.

[0181] The parameters of the recommendation generation model, such as the learning rate and regularization coefficient, can be optimized according to the hidden state vector. The specific formula for iteratively updating the model parameters is shown in Equation (10).

[0182] (10)

[0183] Among them, z t represents the update gate, r t represents the input at the current moment, represents the hidden state at the previous moment, h t represents the hidden state at the current moment, Wz, W h represents the weight matrix, represents the Sigmoid function, and tanh represents the hyperbolic tangent function.

[0184] The method for generating wiring rectification suggestions for power metering devices provided in this embodiment realizes automated and intelligent wiring troubleshooting and rectification through multi-modal data fusion, deep learning models, and online feedback optimization. Compared with manual troubleshooting, the method for generating wiring rectification suggestions for power metering devices provided in this embodiment significantly improves efficiency and reduces costs. The method for generating wiring rectification suggestions for power metering devices provided in this embodiment also improves the accuracy of fault diagnosis and the professionalism and operability of rectification suggestions through a deep model and a constraint generation mechanism. The method for generating wiring rectification suggestions for power metering devices provided in this embodiment also utilizes an online feedback mechanism to achieve continuous optimization of the model, enhance the adaptability to the on-site environment, and solve problems such as low efficiency, reliance on manual experience, lack of professionalism, and lack of continuous optimization ability in the prior art.

[0185] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown sequentially according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps is not strictly limited in order, 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. These steps or stages do not necessarily have to be executed at the same time, but can be executed at different times. The execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or alternately with at least some of the steps or steps in other steps.

[0186] Based on the same inventive concept, the embodiment of the present application also provides a device for generating wiring rectification suggestions for power metering devices for implementing the method for generating wiring rectification suggestions for power metering devices involved above. The implementation solutions for solving problems provided by this device are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the device for generating wiring rectification suggestions for power metering devices provided below can refer to the limitations on the method for generating wiring rectification suggestions for power metering devices in the above text, and will not be repeated here.

[0187] In an exemplary embodiment, as Figure 4 shown, a device for generating wiring rectification suggestions for power metering devices is provided, where:

[0188] The wiring fault data acquisition module 401 is configured to acquire historical wiring fault text data, historical wiring fault image data, and historical wiring fault time series data of the power metering device;

[0189] The wiring fault feature extraction module 402 is used to extract features from the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features;

[0190] The wiring fault feature extraction module 402 is further used to inject position information into the historical wiring fault time series data, and extract the correlation information of the historical wiring fault time series data after injecting the position information according to the self-attention mechanism in the transformer encoder, so as to obtain historical wiring fault time series features;

[0191] The feature fusion module 403 is used to fuse the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features to obtain cross-modal fusion features;

[0192] The wiring rectification suggestion generation module 404 is used to obtain the wiring rectification suggestions for the power metering equipment according to the cross-modal fusion features and the suggestion generation model.

[0193] In one embodiment, the wiring fault feature extraction module 402 is further used to: extract historical wiring fault text features from the historical wiring fault text data according to the bidirectional encoder representation model based on the transformer; extract historical wiring fault image features from the historical wiring fault image data according to the dense connection network model.

[0194] In one embodiment, the feature fusion module 403 is further used to: calculate the respective weights of the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features according to the gating mechanism; perform weighted fusion on the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features according to the respective weights of the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features to obtain cross-modal fusion features.

[0195] In one embodiment, the wiring rectification suggestion generation module 404 is further used to: input the cross-modal fusion features into the suggestion generation model to obtain candidate wiring rectification suggestions; check whether each candidate word in the candidate wiring rectification suggestions belongs to the pre-constructed power term dictionary; when there is a candidate word that does not belong to the power term dictionary, remove the candidate word from the candidate wiring rectification suggestions, or replace the candidate word with a word with similar semantics in the power term dictionary to obtain the wiring rectification suggestions for the power metering equipment.

[0196] In one embodiment, the device further includes a model optimization module, configured to: obtain a feedback vector according to user feedback in the actual operation and maintenance process of wiring rectification of the power metering device; input the feedback vector into a gated recurrent unit model to obtain a hidden state vector; and optimize the recommendation generation model according to the hidden state vector.

[0197] In one embodiment, the historical wiring fault text data of the power metering device includes alarm logs, manual inspection records, and work order system data; the historical wiring fault image data of the power metering device includes on-site photos of wiring errors; and the historical wiring fault time series data of the power metering device includes historical error occurrence frequencies and device status time series metrics.

[0198] Each module in the above device for generating wiring rectification recommendations for the power 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 of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above respective modules.

[0199] 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 a 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 data of embodiments of the method for generating wiring rectification recommendations for the power metering device. 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 a method for generating wiring rectification recommendations for the power metering device.

[0200] Those skilled in the art can understand that Figure 5 the structure shown in

[0201] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the foregoing method embodiments are implemented.

[0202] In one 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 steps in the foregoing method embodiments are implemented.

[0203] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the foregoing method embodiments are implemented.

[0204] 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.

[0205] 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 this 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 this 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 this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0206] 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 this application.

[0207] 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 to 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 generating wiring rectification suggestions for electric energy metering equipment, characterized in that: The method comprises: Acquire historical wiring fault text data, historical wiring fault image data and historical wiring fault time series data of electric energy metering equipment; Performing feature extraction on the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features; injecting position information into the historical wiring fault time series data, and extracting the associated information of the historical wiring fault time series data after the position information is injected according to the self-attention mechanism in the converter encoder, so as to obtain the historical wiring fault time series features; Performing feature fusion on the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain a cross-modal fusion feature; According to the cross-modal fusion features and suggestion generation model, wiring rectification suggestions for the electric energy metering device are obtained.

2. The method according to claim 1, characterized in that: The feature extraction of the historical wiring fault text data and the historical wiring fault image data to obtain the historical wiring fault text features and the historical wiring fault image features includes: According to the dense connection network model of the transformer-based bidirectional encoder representation model, feature extraction is performed on the historical wiring fault text data to obtain historical wiring fault text features; According to the densely connected network model, feature extraction is performed on the historical wiring fault image data to obtain historical wiring fault image features.

3. The method according to claim 1, characterized in that The step of fusing the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features to obtain cross-modal fusion features includes: Calculate the weights corresponding to the historical wiring fault text features, the historical wiring fault image features, and the historical wiring fault time series features according to the gating mechanism; According to the weights corresponding to the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features, the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features are weightedly fused to obtain a cross-modal fusion feature.

4. The method according to claim 1, characterized in that The step of obtaining wiring rectification suggestions for the electric energy metering device based on the cross-modal fusion features and suggestion generation model includes: Inputting the cross-modal fusion features into a suggestion generation model to obtain candidate wiring rectification suggestions; According to a pre-constructed electric power term dictionary, checking whether each candidate word in the candidate wiring rectification suggestion belongs to the electric power term dictionary; When there is a candidate word that does not belong to the electric power term dictionary, the candidate word is removed from the candidate wiring rectification suggestion, or the candidate word is replaced with a semantically similar word in the electric power term dictionary to obtain the wiring rectification suggestion for the electric energy metering device.

5. The method according to claim 4, characterized in that After obtaining the wiring rectification suggestion of the electric energy metering device, the method further includes: According to the user feedback during the actual operation and maintenance process of the wiring rectification of the electric energy metering equipment, the feedback vector is obtained; Inputting the feedback vector into a gated recurrent unit model to obtain a hidden state vector; The suggestion generation model is optimized based on the hidden state vector.

6. The method according to claim 1, characterized in that The historical wiring fault text data of the electric energy metering device includes alarm logs, manual inspection records and work order system data; the historical wiring fault image data of the electric energy metering device includes on-site photos of wiring errors; the historical wiring fault timing data of the electric energy metering device includes historical error occurrence frequency and equipment status timing indicators.

7. A device for generating wiring rectification suggestions for electric energy metering equipment, characterized in that: The device comprises: A wiring fault data acquisition module is used to acquire historical wiring fault text data, historical wiring fault image data and historical wiring fault time series data of electric energy metering equipment; A wiring fault feature extraction module, used to extract features from the historical wiring fault text data and the historical wiring fault image data to obtain historical wiring fault text features and historical wiring fault image features; The wiring fault feature extraction module is further used to inject position information into the historical wiring fault time series data, and extract the associated information of the historical wiring fault time series data after the position information is injected according to the self-attention mechanism in the converter encoder to obtain the historical wiring fault time series feature; A feature fusion module, used to fuse the historical wiring fault text features, the historical wiring fault image features and the historical wiring fault time series features to obtain a cross-modal fusion feature; A wiring rectification suggestion generation module is used to obtain wiring rectification suggestions for the electric energy metering device according to the cross-modal fusion features and the suggestion generation model.

8. 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 6 are implemented.

9. 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 6 are implemented.

10. A computer program product, comprising a computer program, 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 6 are implemented.

Images