Emergency command disposal multi-source data fusion method based on artificial intelligence

Abstract

The invention belongs to the technical field of specific data processing, and particularly relates to an emergency command disposal multi-source data fusion method based on artificial intelligence. Comprising the following steps: establishing a connection channel with a provincial side power grid resource service platform to obtain multi-source data; uniformly converting data formats of different data sources into standard formats; a dynamic auto-encoder method in deep learning is utilized to automatically learn data potential feature representations and fuse the data potential feature representations into feature vectors; and finally, displaying the fused data in a power grid system and designing a friendly user interface. The feature fusion is realized through the steps of adaptive normalization, dynamic adjustment of an auto-encoder structure, training of an auto-encoder and a discriminator, calculation of an attention weight, introduction of mutual information evaluation and the like. The method can effectively solve the problems of data format difference, inaccurate feature extraction and the like, improves the data fusion accuracy and efficiency, ensures that the data is well displayed in a system, and provides powerful support for power grid emergency command and disposal.

Description

Technical Field

[0001] The present invention belongs to the specific field of data processing technology, and in particular relates to an artificial intelligence-based emergency command and disposal multi-source data fusion method. Background Art

[0002] In the power grid system, the provincial power grid resource business center needs to obtain 31 types of data, including power grid characteristic data, power outage and restoration information, power outage topology analysis, power grid events, equipment ledgers, etc. Among them, the power grid resource business center obtains distribution network switch position change data from the distribution automation system; obtains distribution power outage and power-on data from the power consumption information collection system; obtains main network switch position change data, equipment anomalies, equipment operation exceeding limits, fault actions and other event data from the control cloud.

[0003] However, these data from different systems have different formats, and the data volume is large, complex and diverse, which brings challenges to the display of data in the system, involving data access and fusion issues. At present, when facing these multi-source data, the existing data processing methods have problems such as incomplete data cleaning, difficult format conversion, inaccurate feature extraction, and unsatisfactory fusion effect, which makes it difficult to display the data intuitively and clearly in the system, affecting the user's rapid viewing and analysis of the data, and unable to meet the power grid system's needs for emergency command and disposal. Therefore, this patent aims to propose an artificial intelligence-based emergency command and disposal multi-source data fusion method to effectively solve these problems, improve the accuracy and efficiency of data fusion, ensure that the data can be well displayed in the system, and provide strong support for the emergency command and disposal of the power grid. Summary of the invention

[0004] In view of the technical problems existing in the above-mentioned power grid data processing, the present invention proposes an emergency command and disposal multi-source data fusion method based on artificial intelligence.

[0005] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0006] S1. First, establish a connection channel with the provincial power grid resource business middle station, and obtain multi-source data from the business middle station;

[0007] S2. Secondly, the data formats of different data sources are uniformly converted into a standard format that can be recognized and processed by the power grid system, so as to facilitate subsequent data fusion and display;

[0008] S3, then use the dynamic autoencoder method in deep learning to automatically learn the potential feature representations of data from different data sources and fuse them into a comprehensive feature vector;

[0009] S4. Finally, the fused data is displayed in the power grid system in an intuitive and clear manner, and a friendly user interface is designed to facilitate users to quickly view and analyze the fused data.

[0010] In step S3, the dynamic autoencoder method in deep learning is used to automatically learn the potential feature representation of data from different data sources and fuse them into a comprehensive feature vector as follows:

[0011] S31. First, an adaptive normalization method is used to dynamically normalize the data according to the mean and standard deviation of each data source. For a data point x of data source i, the normalization operation is: where μ i ,σ i are the mean and standard deviation of the data from data source i respectively;

[0012] S32. Secondly, during the training process, the structure of the autoencoder is dynamically selected and adjusted according to the characteristics of the input data. The optimal module combination is selected for the next step of training by evaluating the performance indicators of different module combinations on the current data batch. The adjustment of the autoencoder structure is regarded as a decision-making process using the policy gradient algorithm, and the selection of structural adjustments that can improve the fusion effect is encouraged through the reward mechanism.

[0013] S33, then train the autoencoder and the discriminator at the same time. For the input data X, the autoencoder generates a fusion feature F, and the probability that the discriminator judges that F is the real data is D(F). The loss function of the autoencoder includes reconstruction loss and adversarial loss, that is, L autoencoder =L reconstruction +λL adversarial , where L reconstruction is the reconstruction loss, L adversarial = -log(D(F)) is the adversarial loss, and λ is the weight parameter that balances the two losses;

[0014] S34, then calculate the attention weights of different data source features, and for each data source, the potential feature representation Z i , calculate its attention weight α through an attention network i , the fused feature vector is Where k is the number of data sources;

[0015] S35. Finally, mutual information is introduced as an evaluation indicator to measure the quality of fusion features. According to the evaluation results, the parameters and fusion strategy of the autoencoder are further optimized.

[0016] Preferably, the data obtained from the business middle station in step S1 includes power grid characteristic data, power outage and restoration information, power outage topology analysis, power grid events, and equipment inventory data.

[0017] Preferably, the specific implementation of the unified data format conversion in step S2 is:

[0018] S21. First, we deeply analyze the data requirements of the power grid system, determine a standard data format that can be efficiently recognized and processed by the power grid system, and collect various types of sample data from different data sources for training the adaptive conversion engine;

[0019] S22. Next, for each type of data sample, extract its key features. For the table data T = {C1, C2, ... C n}, where C i represents the column, and the extracted features are represented as F(T)={f1(T),f1(T),...,f m (T)}, where f j (T) represents the jth feature extraction function;

[0020] S23, then use the decision tree to learn the extracted features and establish the conversion rules from different data source formats to the standard format. The input data feature is X = {x1, x2, ... x k}, the output standard format is Y, so that the corresponding standard format output can be accurately predicted for a given input feature;

[0021] S24, then when new data needs to be formatted, feature extraction is performed on the new data, and the features extracted in the training phase are matched to find the most similar sample data type, thereby determining the applicable conversion rules. new The extracted features are F(D new )={f1(D new ),f2(D new ),...,f m (D new )}, then calculate the similarity with the sample data features and find the most similar sample data type T similar ;

[0022] S25. According to the determined conversion rules, the new data is converted into a format, and the data from different data sources are converted into a standard format that can be recognized and processed by the power grid system;

[0023] S26. Finally, the converted results are verified by randomly selecting part of the data for manual inspection to check whether they meet the requirements of the standard format. If the verification fails, the rules are adjusted and the conversion and verification are repeated.

[0024] Preferably, the design of a user-friendly interface involved in step S4 includes taking into account the screen size and resolution of different devices, and adopting a responsive design to ensure that the user interface can be well displayed on various devices; integrating some data analysis tools into the user interface including data statistics functions, supporting data export functions and integrated predictive analysis functions, so as to facilitate users to conduct further analysis and decision-making.

[0025] Compared with the prior art, the advantages and positive effects of the present invention are that the dynamic autoencoder method can automatically learn the potential feature representation of data from different data sources, and optimize it through adaptive normalization, policy gradient algorithm and other technologies, thereby improving the accuracy of data fusion. By deeply analyzing the requirements of the power grid system for data, using adaptive conversion engines and decision tree learning and other technologies, the data formats of different data sources can be quickly and accurately converted into a standard format that the power grid system can recognize and process, thereby improving data processing efficiency. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0028] Embodiment: In the power grid system, it is necessary to fuse and analyze multi-source data from different data sources to support emergency command and disposal decisions. However, the data formats of different data sources are different, and the data volume is huge, complex and diverse, which brings challenges to the fusion and display of data. The present invention provides an artificial intelligence-based emergency command and disposal multi-source data fusion method, which can effectively solve these problems, improve the accuracy and efficiency of data fusion, and provide strong support for the emergency command and disposal of the power grid.

[0029] First, in order to obtain real power grid data, it is necessary to establish a connection channel with the provincial power grid resource business platform, and obtain multi-source data from the business platform. Multi-source data includes power grid characteristic data, power outage and restoration information, power outage topology analysis, power grid events, and equipment ledger data. The collection of multi-source data provides a basis for subsequent data processing and analysis. By obtaining these rich data, we can have a more comprehensive understanding of the operating status and potential problems of the power grid.

[0030] Next, since the data formats of different data sources may be different, the data needs to be converted to ensure data consistency and accuracy. First, we deeply analyze the data requirements of the power grid system and determine a standard data format that can be efficiently recognized and processed by the power grid system. We collect various types of sample data from different data sources to train the adaptive conversion engine. Secondly, for each type of data sample, we extract its key features. For the table data T = {C1, C2, ... C n}, where C i represents the column, and the extracted features are represented as F(T)={f1(T),f1(T),...,f m (T)}, where f j (T) represents the jth feature extraction function; then the decision tree is used to learn the extracted features and establish the conversion rules from different data source formats to the standard format. The input data feature is X = {x1, x2, ... x k}, the output standard format is Y, so that the corresponding standard format output can be accurately predicted for the given input features; then when new data needs to be converted, the new data is extracted and matched with the features extracted in the training phase to find the most similar sample data type, thereby determining the applicable conversion rules. new The extracted features are F(D new )={f1(D new ),f2(D new ),...,f m (D new )}, then calculate the similarity with the sample data features and find the most similar sample data type T similar ; According to the determined conversion rules, the new data is converted into a standard format that can be recognized and processed by the power grid system; finally, the converted results are verified by randomly selecting some data for manual inspection to check whether they meet the requirements of the standard format. If the verification fails, the rules are adjusted and the conversion and verification are performed again. This data conversion uses technologies such as adaptive conversion engines and decision tree learning, which can quickly and accurately convert formats, improve the efficiency of data processing, and save time and resources. It enables the power grid system to be better compatible with data from different data sources, breaks the barriers of data formats, and promotes the integration and sharing of multi-source data.

[0031] In order to effectively capture the complex nonlinear relationship between data from different data sources, dig out the hidden patterns and information in the data, and provide richer and more representative feature representation for data fusion. The present invention uses the dynamic encoder method in data fusion. First, an adaptive normalization method is used to perform dynamic normalization according to the mean and standard deviation of each data source data. For data point x of data source i, the normalization operation is where μ i ,σ i are the mean and standard deviation of the data from data source i respectively; secondly, during the training process, the structure of the autoencoder is dynamically selected and adjusted according to the characteristics of the input data, and the optimal module combination is selected for the next step of training by evaluating the performance indicators of different module combinations on the current data batch. The policy gradient algorithm is used to regard the adjustment of the autoencoder structure as a decision-making process, and the reward mechanism is used to encourage the selection of structural adjustments that can improve the fusion effect; then the autoencoder and the discriminator are trained simultaneously. For the input data X, the autoencoder generates a fusion feature F, and the probability that the discriminator judges that F is the real data is D(F). The loss function of the autoencoder includes reconstruction loss and adversarial loss, that is, L autoencoder =L reconstruction +λL adversarial , where L reconstruction is the reconstruction loss, L adversarial = -log(D(F)) is the adversarial loss, λ is the weight parameter to balance the two losses; then the attention weights of the features of different data sources are calculated, and the potential feature representation Z for each data source is i , calculate its attention weight α through an attention network i , the fused feature vector is Where k is the number of data sources; finally, mutual information is introduced as an evaluation indicator to measure the quality of fusion features. According to the evaluation results, the parameters and fusion strategy of the autoencoder are further optimized.

[0032] Finally, the fused data is displayed in an intuitive and clear way in the power grid system. By designing a friendly user interface, users can quickly view and analyze the fused data. Considering the screen size and resolution of different devices, a responsive design is adopted to ensure that the user interface can be well displayed on various devices. Media query technology is used in the system to detect the screen size and resolution of different devices, and fluid layout and relative units are used to ensure that the overall layout and font size and element size are displayed consistently on different devices. For data analysis tools, including data statistics functions, support for data export functions and integrated predictive analysis functions, the Pandas library in Python is used for data statistical calculations, and the Pandas.to_csv() function is used to export data to CSV format, providing strong support for the emergency command and disposal of the power grid.

[0033] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A multi-source data fusion method for emergency command and disposal based on artificial intelligence, characterized in that: The following steps are involved: S1. First, establish a connection channel with the provincial power grid resource business middle station, and obtain multi-source data from the business middle station; S2. Secondly, the data formats of different data sources are uniformly converted into a standard format that can be recognized and processed by the power grid system, so as to facilitate subsequent data fusion and display; S3, then use the dynamic autoencoder method in deep learning to automatically learn the potential feature representations of data from different data sources and fuse them into a comprehensive feature vector; S4. Finally, the fused data is displayed in an intuitive and clear way in the power grid system, and a friendly user interface is designed to facilitate users to quickly view and analyze the fused data; In step S3, the dynamic autoencoder method in deep learning is used to automatically learn the potential feature representation of data from different data sources and fuse them into a comprehensive feature vector as follows: S31. First, an adaptive normalization method is used to dynamically normalize the data according to the mean and standard deviation of each data source. For a data point x of data source i, the normalization operation is: where μ i ,σ i are the mean and standard deviation of the data from data source i; S32. Secondly, during the training process, the structure of the autoencoder is dynamically selected and adjusted according to the characteristics of the input data. The optimal module combination is selected for the next step of training by evaluating the performance indicators of different module combinations on the current data batch. The adjustment of the autoencoder structure is regarded as a decision-making process using the policy gradient algorithm, and the selection of structural adjustments that can improve the fusion effect is encouraged through the reward mechanism. S33, then train the autoencoder and the discriminator at the same time. For the input data X, the autoencoder generates a fusion feature F, and the probability that the discriminator judges that F is the real data is D(F). The loss function of the autoencoder includes reconstruction loss and adversarial loss, that is, L autoencoder =L reconstruction +λL adversarial , where L reconstruction is the reconstruction loss, L adversarial = -log(D(F)) is the adversarial loss, and λ is the weight parameter that balances the two losses; S34, then calculate the attention weights of different data source features, and represent the potential features Z of each data source i , calculate its attention weight α through an attention network i , the fused feature vector is Where k is the number of data sources; S35. Finally, mutual information is introduced as an evaluation indicator to measure the quality of fusion features. According to the evaluation results, the parameters and fusion strategy of the autoencoder are further optimized.

2. According to the artificial intelligence-based emergency command and disposal multi-source data fusion method of claim 1, it is characterized in that: The data obtained from the business platform in step S1 includes power grid characteristic data, power outage and restoration information, power outage topology analysis, power grid events, and equipment inventory data.

3. According to the artificial intelligence-based emergency command and disposal multi-source data fusion method of claim 1, it is characterized in that: The specific implementation of the unified data format conversion in step S2 is: S21. First, we deeply analyze the data requirements of the power grid system, determine a standard data format that can be efficiently recognized and processed by the power grid system, and collect various types of sample data from different data sources for training the adaptive conversion engine; S22. Secondly, for each type of data sample, extract its key features. Table data T = {C1, C2, ... C n }, where C i represents the column, and the extracted features are represented as F(T)={f1(T),f1(T),...,f m (T)}, where f j (T) represents the jth feature extraction function; S23, then use the decision tree to learn the extracted features and establish the conversion rules from different data source formats to the standard format. The input data feature is X = {x1, x2, ... x k }, the output standard format is Y, so that the corresponding standard format output can be accurately predicted for a given input feature; S24, then when new data needs to be formatted, feature extraction is performed on the new data, and the features extracted in the training phase are matched to find the most similar sample data type, thereby determining the applicable conversion rules. new The extracted features are F(D new )={f1(D new ),f2(D new ),...,f m (D new )}, then calculate the similarity with the sample data features and find the most similar sample data type T similar ; S25. According to the determined conversion rules, the new data is converted into a format, and the data from different data sources are converted into a standard format that can be recognized and processed by the power grid system; S26. Finally, the converted results are verified by randomly selecting part of the data for manual inspection to check whether they meet the requirements of the standard format. If the verification fails, the rules are adjusted and the conversion and verification are repeated.

4. According to the artificial intelligence-based emergency command and disposal multi-source data fusion method of claim 1, it is characterized in that: The design of a user-friendly interface involved in step S4 includes taking into account the screen size and resolution of different devices, and adopting a responsive design to ensure that the user interface can be well displayed on various devices; integrating some data analysis tools into the user interface, including data statistics functions, supporting data export functions and integrated predictive analysis functions, to facilitate users to conduct further analysis and decision-making.

Images