Intelligent electric quantity abnormity monitoring and missing data complementing system

Abstract

The invention relates to the technical field of electric power, and discloses an intelligent electric quantity abnormity monitoring and missing data complementing system. The intelligent electric quantity abnormity monitoring and missing data complementing system comprises a data preprocessing module, a customized attention model and a prediction and processing module, and through a customized attention mechanism, the model can more accurately capture a complex relationship among data, especially dependency among cross-tables and fields, so that the data can be more accurately monitored. Therefore, the accuracy and reliability of missing value prediction are improved; the model is combined with business knowledge and statistical information, and a prediction result is more in line with an actual business scene through field relation graph construction and attention weight adjustment; when the model is used for processing large-scale multi-table data, high processing speed and efficiency can be kept, calculation cost and time cost are reduced, and the technical scheme has obvious advantages in the aspects of improving prediction precision, enhancing service adaptability and optimizing calculation efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and specifically relates to an intelligent power anomaly monitoring and missing data completion system. Background Art

[0002] In the electricity market transaction, the existing technology has proposed a series of data fitting rules to solve the problem of missing data caused by the failure of meter data collection. These rules include taking different strategies for fitting according to the different situations of the number of defects in continuous time points. For example, when the number of defects is small, the arithmetic mean of the interval electricity quantity of the previous and next time points is used for fitting; when the number of defects is large, the corresponding historical data is selected according to the date attribute for approximate fitting. These technologies provide an effective solution for supplementing the missing data when the meter fails to collect.

[0003] In the field of power collection, when existing power collection terminals read meter data, some data is often missed due to network fluctuations, equipment failures and other reasons, especially the missed reading of key information such as daily frozen data and curve data, which has a serious impact on power grid monitoring and user assessment. Traditional methods mostly rely on manual verification and manual re-entry, which is not only inefficient but also prone to errors. In recent years, with the development of broadband carrier technology, the success rate of automatic meter reading of power collection terminals has been significantly improved, but there is still a problem of missed reading due to the inability to collect curve data during the reading period of daily frozen data. Therefore, an efficient and automated data re-reading technology is needed to ensure that the power collection terminal can completely and accurately obtain all meter data, so as to meet the needs of power grid operation and management. The present invention is proposed to solve this technical problem. By referring to the association between other fields and missing fields, the automatic completion of data is realized through intelligent algorithms, the data integrity and accuracy are improved, and the efficiency and intelligence level of power grid operation and management are improved. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent power anomaly monitoring and missing data completion system, which effectively solves the problem in the field of power collection that when the existing power collection terminals read the meter data, some data are often missed due to network fluctuations, equipment failures, etc., especially the omission of key information such as daily frozen data and curve data, which has a serious impact on power grid monitoring and user assessment. Traditional methods mostly rely on manual verification and manual supplementation, which is not only inefficient but also prone to errors. In recent years, with the development of broadband carrier technology, the success rate of automatic meter reading of power collection terminals has been significantly improved, but there is still a problem of missed reading due to the inability to collect curve data during the reading period of daily frozen data.

[0005] To achieve the above object, the present invention provides the following technical solutions: an intelligent power anomaly monitoring and missing data completion system, the intelligent power anomaly monitoring and missing data completion system comprising a data preprocessing module, a customized attention model, and a prediction and processing module;

[0006] Data preprocessing module: The data preprocessing module includes a data cleaner, a serializer, and a table embedding generator; the data cleaner is responsible for identifying and processing abnormal values ​​and default value pre-filling and data normalization in the input multi-table data. The pre-filling method of missing values ​​includes filling missing values ​​with zero, mean, median, mode, or combining specific business needs; the serializer sorts and organizes the data according to preset rules; the table embedding generator is responsible for capturing table-level information, and the system generates a unique embedding vector through the table;

[0007] Customized attention model: The customized attention model includes a basic attention layer and a table relationship enhancement layer; the basic attention layer uses a self-attention mechanism to process serialized data, and the self-attention mechanism can capture the complex dependencies between elements within the sequence; the table relationship enhancement layer includes a field relationship graph builder, an attention weight adjuster, and a cross-table attention mechanism; the field relationship graph builder is based on business knowledge and statistical information, and constructs a field relationship graph with fields as nodes and dependencies or correlations between fields as edges; the attention weight adjuster uses table embedding and field relationship graph to adjust the attention weights calculated by the basic attention layer; the cross-table attention mechanism is used to allow the model to consider information from different tables when calculating attention weights;

[0008] Prediction and processing module: The prediction and processing module includes a prediction network and a processing module; the prediction network is responsible for using the enhanced attention weights and other field values ​​in the sequence to predict the value of the missing field, and the prediction network adopts a fully connected layer deep learning architecture; the processing module includes evaluation indicators and output formatters, and the evaluation indicators are used to measure the degree of difference between the model prediction results and the actual results, as well as the overall performance of the model; the output formatter is used to ensure that the prediction results can be seamlessly integrated into the existing data system; the formatter performs data cleaning and organization as needed.

[0009] Preferably, the outlier detection in the data preprocessing module adopts the Z-score statistical method, and the Z-score calculation formula is: The data preprocessing formula of the data cleaner is:

[0010] Preferably, the basic attention layer adopts the multi-head attention mechanism in the Transformer model, and the self-attention calculation formula of each group of independent heads in the multi-head attention mechanism is: Multi-head attention performs multiple independent attention operations in parallel and concatenates the results. The concatenation formula is MultiHead(Q, K, V) = Concat(head1, ..., head h )W 0 The attention weight adjuster uses table embedding and field relationship graph to adjust the attention weight generated by the basic attention layer. The adjusted attention weight is expressed as W adj =σ(W atta +Weightfunction(A, T)); The cross-table attention mechanism is implemented by introducing cross-table embedding or relationship matrix to allow the model to consider information from different tables when calculating attention weights. The cross-table attention weights are designed as

[0011] Preferably, the prediction network in the prediction and processing module adopts a multi-layer fully connected neural network, and the multi-layer fully connected neural network includes an input layer, a hidden layer and an output layer.

[0012] Preferably, the input layer is used in the prediction process, ignoring the pre-filled default value, and the prediction network adopts a mask mechanism. The input layer is used to receive the feature vector processed by the customized attention model, which includes the serialized data field value and the mask. The input layer formula is

[0013] Preferably, the hidden layer adopts a multi-layer fully connected neural network structure, and each layer of the fully connected neural network increases the nonlinear ability of the model through a nonlinear activation function, which is expressed as h l =ReLU(W l h l-1 +b l ).

[0014] Preferably, the output layer is designed according to the specific task, and its output prediction value formula is:

[0015] Preferably, the processing module is responsible for evaluating the quality of the prediction results and preparing output data for subsequent use, and uses the evaluation indicators MSE, MAE and RMSE to measure the accuracy of the model prediction, where the formula of MSE is The formula for MAE is The formula for RMSE is

[0016] Preferably, the output formatter is used to organize the prediction results into a format that meets business needs, including filling the prediction values ​​back into the corresponding positions of the original data table, performing necessary data cleaning and organization work, generating reports or visual charts, and integrating the prediction results into the data system.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention, through a customized attention mechanism, enables the model to more accurately capture the complex relationships between data, especially the dependencies across tables and fields, thereby improving the accuracy and reliability of missing value prediction;

[0018] 2. The present invention combines the model with business knowledge and statistical information, builds a field relationship graph and adjusts the attention weight, so that the prediction results are more in line with the actual business scenario;

[0019] 3. The present invention, when using the model to process large-scale multi-table data, can maintain a high processing speed and efficiency, reduce computing costs and time costs, and the technical solution shows obvious advantages in improving prediction accuracy, enhancing business adaptability, and optimizing computing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0021] In the attached picture:

[0022] Figure 1 The block diagram of an intelligent power anomaly monitoring and missing data completion system of the present invention. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the present invention provides a technical solution: an intelligent power anomaly monitoring and missing data completion system, the intelligent power anomaly monitoring and missing data completion system includes a data preprocessing module, a customized attention model, and a prediction and processing module;

[0025] Data preprocessing module: The data preprocessing module includes a data cleaner, a serializer, and a table embedding generator; the data cleaner is responsible for identifying and processing abnormal values ​​and default value pre-filling and data normalization in the input multi-table data. The pre-filling method of missing values ​​includes filling missing values ​​with zero, mean, median, mode, or combining specific business needs; the serializer sorts and organizes the data according to preset rules; the table embedding generator is responsible for capturing table-level information, and the system generates a unique embedding vector through the table;

[0026] Customized attention model: The customized attention model includes a basic attention layer and a table relationship enhancement layer; the basic attention layer uses a self-attention mechanism to process serialized data, and the self-attention mechanism can capture the complex dependencies between elements within the sequence; the table relationship enhancement layer includes a field relationship graph builder, an attention weight adjuster, and a cross-table attention mechanism; the field relationship graph builder is based on business knowledge and statistical information, and constructs a field relationship graph with fields as nodes and dependencies or correlations between fields as edges; the attention weight adjuster uses table embedding and field relationship graph to adjust the attention weights calculated by the basic attention layer; the cross-table attention mechanism is used to allow the model to consider information from different tables when calculating attention weights;

[0027] Prediction and processing module: The prediction and processing module includes a prediction network and a processing module; the prediction network is responsible for using the enhanced attention weights and other field values ​​in the sequence to predict the value of the missing field, and the prediction network adopts a fully connected layer deep learning architecture; the processing module includes evaluation indicators and output formatters, and the evaluation indicators are used to measure the degree of difference between the model prediction results and the actual results, as well as the overall performance of the model; the output formatter is used to ensure that the prediction results can be seamlessly integrated into the existing data system; the formatter performs data cleaning and organization as needed.

[0028] Among them, the outlier detection in the data preprocessing module adopts the Z-score statistical method, and the Z-score calculation formula is: The data preprocessing formula of the data cleaner is: The basic attention layer uses the multi-head attention mechanism in the Transformer model. The self-attention calculation formula for each group of independent heads in the multi-head attention mechanism is Multi-head attention performs multiple independent attention operations in parallel and concatenates the results. The concatenation formula is MultiHead(Q, K, V) = Concat(head1, ..., head h )W 0 The attention weight adjuster uses table embedding and field relationship graph to adjust the attention weight generated by the basic attention layer. The adjusted attention weight is expressed as W adj =σ(W atta +Weightfunction(A, T)); The cross-table attention mechanism is implemented by introducing cross-table embedding or relationship matrix to allow the model to consider information from different tables when calculating attention weights. The cross-table attention weights are designed as The prediction network in the prediction and processing module uses a multi-layer fully connected neural network, which includes an input layer, a hidden layer, and an output layer. The input layer is used in the prediction process, ignoring the pre-filled default value. The prediction network uses a mask mechanism. The input layer is used to receive the feature vector processed by the customized attention model, which includes the serialized data field value and the mask. The input layer formula is The hidden layer adopts a multi-layer fully connected neural network structure. Each layer of the fully connected neural network increases the nonlinear ability of the model through a nonlinear activation function, which is expressed as h l =ReLU(W l h l-1 +b l ); the output layer is designed according to the specific task, and its output prediction value formula is The processing module is responsible for evaluating the quality of the prediction results and preparing the output data for subsequent use. The evaluation indicators MSE, MAE and RMSE are used to measure the accuracy of the model prediction. The formula of MSE is: The formula for MAE is The formula for RMSE is The output formatter is used to organize the prediction results into a format that meets business needs, including filling the prediction values ​​back into the corresponding positions of the original data table, performing necessary data cleaning and organization, generating reports or visual charts, and integrating the prediction results into the data system.

[0030] The data preprocessing module can ensure that the subsequent models can receive clean, orderly and structured data input. The data cleaner in the data preprocessing module can identify and process the outliers in the input multi-table data, pre-fill the default values ​​and normalize the data. Outliers may include values ​​outside the reasonable range and obviously erroneous data points. Before the sequence is input into the model, the missing values ​​are pre-filled. The filling methods include filling the missing values ​​with zero, mean, median, mode or combining the specific needs of the business. Data normalization enhances the accuracy and completeness of the data by reducing data redundancy and improving data consistency, thereby improving Query efficiency and ease of data management. The serializer is the cleaned data that needs to be further serialized so that the subsequent model can process it efficiently. The serializer sorts and organizes the data according to preset rules to ensure the consistency and order of the data, which helps the model better understand the relationship and dependency between data. The table embedding generator is to capture table-level information. The system will generate a unique embedding vector for each table. These embedding vectors are obtained through an initialization method based on business knowledge, which can reflect the structure, content, and position and role of the table in multi-table data. These embedding vectors will serve as one of the important inputs of the customized attention model.

[0031] The basic attention layer in the customized attention model uses the self-attention mechanism to process the serialized data. The self-attention mechanism can capture the complex dependencies between the elements within the sequence, which is crucial for understanding the interactions between different fields in the table data. The table relationship enhancement layer includes a field relationship graph builder, an attention weight adjuster, and a cross-table attention mechanism. The field relationship graph builder is based on business knowledge and statistical information. This component builds a field relationship graph with fields as nodes and dependencies or correlations between fields as edges. This graph structure helps the model understand the complex relationships between fields and adjust the attention weights accordingly. The attention weight adjuster uses table embeddings and field relationship graphs to adjust the attention weights calculated by the basic attention layer. By introducing a learnable weight matrix or function, the model can learn how to optimize the distribution of attention weights based on table embeddings and field relationship graphs. In this way, the model can pay more attention to information and relationships related to missing fields when predicting missing values. The cross-table attention mechanism is to further enhance the model's cross-table learning ability. This mechanism allows the model to consider information from different tables when calculating attention weights. By introducing cross-table embedding or relationship information, the model can more comprehensively understand the complex relationships between multi-table data and make more accurate missing value predictions based on this.

[0032] The prediction network in the prediction and processing module is one of the core parts of the system. It is responsible for predicting the value of the missing field using the enhanced attention weights and other field values ​​in the sequence. The prediction network can use a fully connected layer deep learning architecture. In the prediction process, the prediction network will use the mask mechanism to distinguish between non-missing values ​​and default values, so as to achieve the process of predicting missing values, ignoring the pre-filled default values ​​as much as possible, and focusing on non-default values, and using appropriate loss functions for reverse derivation to obtain the most suitable parameters. The processing module includes evaluation indicators and output formatters. The evaluation indicators are used to measure the degree of difference between the model's predicted results and the actual results, as well as the overall performance of the model. These indicators are very important for understanding the accuracy, bias and variance of the model. The output formatter can ensure that the prediction results can be seamlessly integrated into the existing data system for subsequent use and analysis. The formatter can also perform data cleaning and sorting as needed to ensure the quality and consistency of the output data.

[0033] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0034] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent power anomaly monitoring and missing data completion system, characterized by: The intelligent power anomaly monitoring and missing data completion system includes a data preprocessing module, a customized attention model, and a prediction and processing module; Data preprocessing module: The data preprocessing module includes a data cleaner, a serializer and a table embedding generator; The data cleaner is responsible for identifying and processing outliers and default value pre-filling and data normalization in the input multi-table data. The pre-filling methods for missing values ​​include filling missing values ​​with zero, mean, median, mode, or combining specific business needs; the serializer sorts and organizes the data according to preset rules; the table embedding generator is responsible for capturing table-level information, and the system generates unique embedding vectors through the table; Customized attention model: The customized attention model includes a basic attention layer and a table relationship enhancement layer; the basic attention layer uses a self-attention mechanism to process serialized data, and the self-attention mechanism can capture the complex dependencies between elements within the sequence; the table relationship enhancement layer includes a field relationship graph builder, an attention weight adjuster, and a cross-table attention mechanism; the field relationship graph builder is based on business knowledge and statistical information, and constructs a field relationship graph with fields as nodes and dependencies or correlations between fields as edges; The attention weight adjuster uses table embedding and field relationship graph to adjust the attention weights calculated by the base attention layer; The cross-table attention mechanism is used to allow the model to consider information from different tables when calculating attention weights; Prediction and processing module: The prediction and processing module includes a prediction network and a processing module; the prediction network is responsible for predicting the value of the missing field using the enhanced attention weights and other field values ​​in the sequence, and the prediction network adopts a fully connected layer deep learning architecture; the processing module includes evaluation indicators and output formatters, and the evaluation indicators are used to measure the degree of difference between the model prediction results and the actual results, as well as the overall performance of the model; the output formatter is used to ensure that the prediction results can be seamlessly integrated into the existing data system; The formatter cleans and organizes the data as needed.

2. The intelligent power anomaly monitoring and missing data completion system according to claim 1 is characterized in that: The outlier detection in the data preprocessing module adopts the Z-score statistical method, and the Z-score calculation formula is: The data preprocessing formula of the data cleaner is:

3. The intelligent power anomaly monitoring and missing data completion system according to claim 1 is characterized in that: The basic attention layer adopts the multi-head attention mechanism in the Transformer model. The self-attention calculation formula of each group of independent heads in the multi-head attention mechanism is Multi-head attention performs multiple independent attention operations in parallel and concatenates the results. The concatenation formula is MultiHead(Q, K, V) = Concat(head1, ..., head h )W 0 The attention weight adjuster uses table embedding and field relationship graph to adjust the attention weight generated by the basic attention layer. The adjusted attention weight is expressed as W adj =σ(W atta +Weigh tfunction(A,T)); The cross-table attention mechanism is implemented by introducing cross-table embedding or relationship matrix to allow the model to consider information from different tables when calculating attention weights. The cross-table attention weights are designed as 4. The intelligent power anomaly monitoring and missing data completion system according to claim 1 is characterized in that: The prediction network in the prediction and processing module adopts a multi-layer fully connected neural network, and the multi-layer fully connected neural network includes an input layer, a hidden layer and an output layer.

5. The intelligent power anomaly monitoring and missing data supplementation system according to claim 4 is characterized in that: The input layer is used in the prediction process, ignoring the pre-filled default value. The prediction network adopts a mask mechanism. The input layer is used to receive the feature vector processed by the customized attention model, which includes the serialized data field value and the mask. The input layer formula is 6. The intelligent power anomaly monitoring and missing data completion system according to claim 5 is characterized in that: The hidden layer adopts a multi-layer fully connected neural network structure. Each layer of the fully connected neural network increases the nonlinear ability of the model through a nonlinear activation function, which is expressed as h l =ReLU(W l h l-1 +b l ).

7. The intelligent power anomaly monitoring and missing data supplementation system according to claim 6 is characterized by: The output layer is designed according to the specific task, and its output prediction value formula is:

8. The intelligent power anomaly monitoring and missing data supplementation system according to claim 1 is characterized by: The processing module is responsible for evaluating the quality of the prediction results and preparing output data for subsequent use. The evaluation indicators MSE, MAE and RMSE are used to measure the accuracy of the model prediction, where the formula for MSE is: The formula for MAE is The formula for RMSE is 9. The intelligent power anomaly monitoring and missing data supplementation system according to claim 1 is characterized by: The output formatter is used to organize the prediction results into a format that meets business needs, including filling the prediction values ​​back into the corresponding positions of the original data table, performing necessary data cleaning and organization work, generating reports or visual charts, and integrating the prediction results into the data system.

Images