Power grid safety remote monitoring management system based on artificial intelligence

Abstract

The invention relates to the technical field of power system monitoring, in particular to a power grid safety remote monitoring and management system based on artificial intelligence, which comprises a data acquisition and transmission module, a data processing and analysis module, a fault diagnosis and early warning module, a scheduling control module, a user interaction module and a maintenance support module, according to the scheme, the time relative attention CNN network model is adopted to analyze power grid safety, feature extraction and recognition are carried out on the power grid operation state through time change, dynamic features in the power grid operation state are effectively captured, and monitoring of the power grid operation state is achieved; a fault diagnosis model is constructed by adopting a rapid global self-attention network method based on U-Net, fine features in power grid operation data can be captured from different scales through a multi-scale feature extraction strategy, high calculation cost and memory overhead required for calculating a large attention matrix in a traditional self-attention mechanism are reduced, and the calculation process is accelerated.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to an artificial intelligence-based power grid security remote monitoring and management system. Background Art

[0002] With the development of social economy, people have put forward higher requirements for the quality and reliability of power supply. The rapid development of technologies such as the Internet of Things, big data, cloud computing and artificial intelligence has provided new possibilities for the intelligent management of power grids. However, the existing power grid monitoring system still has obvious deficiencies in many aspects, which restricts its full effectiveness. Many existing monitoring systems still rely on traditional statistical analysis methods and lack support for big data and advanced algorithms, which makes it difficult for the system to deeply explore the value behind the data and cannot achieve comprehensive and accurate monitoring of the power grid operation status. The existing fault diagnosis model is limited in its effectiveness in identifying and locating complex faults, and requires high computing costs and memory consumption. Especially in resource-constrained environments, these problems are prone to response delays. In practical applications, such delays may not only cause serious safety accidents, but also cause significant economic losses and negative social impacts. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a remote monitoring and management system for power grid security based on artificial intelligence. In view of the fact that the existing monitoring system still relies on traditional statistical analysis methods and lacks support for big data and advanced algorithms, which makes it difficult for the system to deeply explore the value behind the data, a time relative attention CNN network model is used to analyze power grid security, and the power grid operation status is extracted and identified through time changes, so as to effectively capture the dynamic characteristics in the power grid operation status and realize comprehensive and accurate monitoring of the power grid operation status; in view of the fact that the existing fault diagnosis model has limited performance in identifying and locating complex faults, and requires high computing costs and memory consumption, a fault diagnosis model is constructed by a method based on a fast global self-attention network of U-Net. The method adopts a multi-scale feature extraction strategy, combined with patch embedding, patch merging and jump connection technology, which can capture subtle features in power grid operation data from different scales, effectively integrate local and global information, and enhance the recognition accuracy of the target area. The fast global self-attention mechanism is used to reduce the high computing cost and memory overhead required for calculating large attention matrices in the traditional self-attention mechanism, accelerate the calculation process, and greatly reduce memory usage, so that the model can run efficiently in a resource-constrained environment.

[0004] The invention provides an artificial intelligence-based remote monitoring and management system for power grid security, which includes a data acquisition and transmission module, a data processing and analysis module, a fault diagnosis and early warning module, a dispatching and control module, a user interaction module, and a maintenance support module;

[0005] The data acquisition and transmission module uses sensors and smart meter devices to collect data from the power grid, and transmits the data to the data processing and analysis module through wired and wireless networks to obtain initial data;

[0006] The data processing and analysis module uses the time relative attention CNN network model to identify changes in initial data, build a power grid operation status assessment model, and generate assessment data;

[0007] The fault diagnosis and early warning module uses a U-Net-based fast global self-attention network method to build a fault diagnosis model based on the evaluation data, identify the existing fault risk points, and achieve early warning of faults. By setting thresholds, it can send an alarm to the operator before the fault occurs and generate a warning report;

[0008] The dispatch control module uses machine learning technology to optimize the grid operation strategy, realize the automatic control function, automatically adjust the load distribution and power generation plan, and generate control records;

[0009] The user interaction module provides a user-friendly graphical interface, displays the data generated by all modules, supports access by multiple terminal devices, and enables users to understand the status of the power grid anytime and anywhere;

[0010] The maintenance support module has a safety protection mechanism, provides daily operation and maintenance services for the system, and provides effective solutions when faced with unexpected problems.

[0011] The data processing and analysis module uses the time relative attention CNN network model to identify changes in initial data, specifically including the following steps:

[0012] Step S1: Data preprocessing, dividing the initial data into continuous sequence data, and preprocessing using one-hot encoding, regularization and standardization methods to obtain preprocessed data;

[0013] Step S2: backbone network design, using the basic structure of the CNN network to build the network, and using the attention mechanism to optimize the CNN network to obtain an improved CNN network;

[0014] Step S3: feature extraction, using the improved CNN network to extract features and obtain high-level features;

[0015] Step S4: feature fusion and classification, adjusting the number of output nodes of the FC layer in the improved CNN network and optimizing the classification of the improved CNN network;

[0016] Step S5: training and testing, dividing the preprocessed data into a training set and a test set in a ratio of 4:1, using the training set to train the improved CNN network, using the test set data for verification, and using the cross entropy loss function and the triplet loss function to evaluate and optimize the improved CNN network;

[0017] Step S6: Hyperparameter tuning, using the Bayesian optimization method to optimize the hyperparameters of the improved CNN network to obtain the final temporal relative attention CNN network model;

[0018] Step S7: network model application, using the final time relative attention CNN network model to build a power grid operation status evaluation model, evaluate the power grid operation status, and obtain evaluation data;

[0019] Furthermore, in step S2, the backbone network design specifically includes the following steps:

[0020] Step S21: Use 7×7 convolution and maximum pooling method to perform downsampling operation to extract low-level features;

[0021] Step S22: Use the average pooling method to reduce the dimension of low-level features, and use the Dropout operation to randomly discard part of the output to extract the overall features;

[0022] Step S23: Use the time series relationship layer to adjust and capture the dynamic changes of power grid data;

[0023] Step S24: self-attention mechanism, using DCSA to dynamically generate convolution kernels, optimizes the learning ability of the CNN network in feature extraction, effectively limits the consumption of calculations and network parameters, and improves the flexibility and efficiency of the CNN network;

[0024] Step S25: Channel and spatial attention mechanism, using the DANet mechanism to allocate attention to the channel dimension and spatial dimension of the CNN network, optimize the feature extraction results, and enhance the CNN network's ability to understand complex scenes;

[0025] Step S26: Normalization and activation, use the GN layer to divide the input preprocessed data into M groups and calculate the mean and variance of each group for normalization. At the same time, use the ReLU activation function to optimize the CNN network.

[0026] The fault diagnosis and early warning module uses a method based on a U-Net fast global self-attention network to build a fault diagnosis model. The method based on a U-Net fast global self-attention network specifically includes the following steps:

[0027] Step M1: construct an encoder and use the U-Net network encoder to extract features from the evaluation data of the fault diagnosis and early warning module to obtain reduced-dimensional feature data;

[0028] Step M2: Fast global self-attention mechanism, using sum normalization to calculate the attention weight, using cyclic multiplication and FFT to calculate the attention matrix, and constructing a fast global self-attention mechanism;

[0029] Step M3: Skip connection: Skip connections are set between different layers of the U-Net network encoder to obtain feature information of different scales.

[0030] Step M4: Construct a decoder, use the U-Net network decoder to process feature information of different scales, and output fault identification results;

[0031] Step M5: Loss function optimization, using binary cross entropy loss and optimization algorithm to train and optimize the U-Net-based fast global self-attention network model to obtain the final fault diagnosis model;

[0032] Further, step M1, constructing an encoder, specifically includes the following steps:

[0033] Step M11: Patch embedding, using a 2×2×2 convolutional layer to perform patch embedding on the evaluation data, converting it into a low-resolution feature space, and outputting embedded feature data;

[0034] Step M12: Transformer self-attention module, using N Transformer self-attention modules to extract deep features from the embedded feature data, wherein the Transformer self-attention module includes layer normalization, a fast global self-attention module, a residual connection, and a multi-layer perceptron;

[0035] Step M13: Patch merging. After the first Transformer self-attention module, the adjacent 2×2×2 convolutional layers are connected using a patch merging layer and processed using a linear layer to obtain reduced-dimensional feature data.

[0036] Furthermore, in step M2, the following steps are specifically included:

[0037] Step M21: Use sum normalization and cyclic multiplication to construct fast global self-attention. The formula used is as follows:

[0038]

[0039] Where Q is the query matrix, K is the key matrix, V is the value matrix, and D K represents the dimension of the key matrix K, Q 2 and K 2represent the squares of the elements in Q and K respectively, is the transposed symbol, Z is a matrix of all 1s, Represents a cyclic multiplication operation, F SA (Q, K, V) represents the features of the fast global self-attention output;

[0040] Step M22: Circular multiplication calculation, the formula used is as follows:

[0041]

[0042] Where IFFT() represents inverse fast Fourier transform, FFT() represents fast Fourier transform, i and j represent the column and row indices of the matrix respectively, C represents the channel dimension, and * represents element-wise multiplication;

[0043] Furthermore, in step M4, a decoder is constructed, which specifically includes the following steps:

[0044] Step M41: feature mapping, using the SERes block to perform feature mapping on the feature information of different scales transmitted by the skip connection to obtain the mapped feature data;

[0045] Step M42: feature upsampling, upsampling the mapped feature data using 3D interpolation to restore it to the size of the original evaluation data, thereby obtaining upsampled feature data;

[0046] Step M43: feature concatenation, concatenating the upsampled feature data with the embedded feature data to form a complete feature map;

[0047] Step M44: Output the result, use a Sigmoid layer to calculate the final fault recognition probability, and output the recognition result.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) In view of the fact that the existing monitoring system still relies on traditional statistical analysis methods and lacks support for big data and advanced algorithms, which makes it difficult for the system to deeply explore the value behind the data, the time relative attention CNN network model is used to analyze the power grid security and perform efficient feature extraction and pattern recognition on the operation status of the power grid according to the changes in the time series. By introducing the consideration of the time dimension, the model can more keenly capture the characteristics of dynamic changes in the operation of the power grid and realize comprehensive and accurate monitoring of the operation status of the power grid;

[0050] (2) In view of the limited effectiveness of existing fault diagnosis models in identifying and locating complex faults and the high computational cost and memory consumption required, a fault diagnosis model is constructed using a U-Net-based fast global self-attention network method. This method adopts a multi-scale feature extraction strategy and combines patch embedding, patch merging and skip connection techniques to capture subtle features in power grid operation data at different scales, effectively integrate local and global information, and enhance the recognition accuracy of the target area. The fast global self-attention mechanism is used to reduce the high computational cost and memory overhead required to calculate large attention matrices in the traditional self-attention mechanism, accelerate the calculation process, and significantly reduce memory usage, so that the model can run efficiently in a resource-constrained environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a power grid security remote monitoring and management system based on artificial intelligence provided by the present invention;

[0052] Figure 2 It is a structural diagram of the method of fast global self-attention network based on U-Net;

[0053] 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. DETAILED DESCRIPTION

[0054] 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.

[0055] Example 1, see Figure 1 , the present invention provides an artificial intelligence-based power grid security remote monitoring and management system, which includes a data acquisition and transmission module, a data processing and analysis module, a fault diagnosis and early warning module, a dispatching and control module, a user interaction module and a maintenance support module;

[0056] The data acquisition and transmission module uses sensors and smart meter devices to collect data from the power grid, and transmits the data to the data processing and analysis module through wired and wireless networks to obtain initial data;

[0057] The data processing and analysis module uses the time relative attention CNN network model to identify changes in initial data, build a power grid operation status assessment model, and generate assessment data;

[0058] The fault diagnosis and early warning module uses a U-Net-based fast global self-attention network method to build a fault diagnosis model based on the evaluation data, identify the existing fault risk points, and achieve early warning of faults. By setting thresholds, it can send an alarm to the operator before the fault occurs and generate a warning report;

[0059] The dispatch control module uses machine learning technology to optimize the grid operation strategy, realize the automatic control function, automatically adjust the load distribution and power generation plan, and generate control records;

[0060] The user interaction module provides a user-friendly graphical interface, displays the data generated by all modules, supports access by multiple terminal devices, and enables users to understand the status of the power grid anytime and anywhere;

[0061] The maintenance support module has a safety protection mechanism, provides daily operation and maintenance services for the system, and provides effective solutions when faced with unexpected problems.

[0062] Embodiment 2, based on the above embodiment, the initial data includes voltage, current, power, frequency and temperature data collected in the substation, transmission line and distribution equipment;

[0063] The generated evaluation data evaluates the degree of voltage fluctuation, current change and equipment operation status based on the voltage, current, power, frequency and temperature data, and comprehensively scores the safety of the grid operation status;

[0064] The identification of existing fault risk points analyzes the fault types and their probability of occurrence under different conditions of the power grid according to the degree of voltage fluctuation, current changes and the operating status of the equipment, and sets a safety threshold. When the safety threshold is exceeded, the early warning mechanism is immediately triggered; for example, if the temperature of the transformer in the substation is detected to be abnormally high, it indicates that the equipment may face the risk of overheating and there are significant safety hazards. At this time, the system will immediately issue an alarm to prompt the power grid staff to quickly take cooling measures to avoid potential faults and ensure the safe and stable operation of the power grid.

[0065] Embodiment three, this embodiment is based on the above embodiment, the user interaction module will display the initial data collected by the data acquisition and transmission module, and update the display in real time. The module will also present the evaluation data generated by the data processing and analysis module, the warning information of the fault diagnosis and warning module, and the control record of the dispatching and control module. By using various chart forms such as bar charts, line charts and pie charts, users can understand the safety status of the power grid more intuitively and clearly, and make decisions.

[0066] Embodiment 4, based on the above embodiment, the data processing and analysis module uses the time relative attention CNN network model to identify the changes of the initial data, specifically including the following steps:

[0067] Step S1: Data preprocessing, dividing the initial data into continuous sequence data, and preprocessing using one-hot encoding, regularization and standardization methods to obtain preprocessed data;

[0068] Step S2: backbone network design, using the basic structure of the CNN network to build the network, and using the attention mechanism to optimize the CNN network to obtain an improved CNN network;

[0069] Step S3: feature extraction, using the improved CNN network to extract features and obtain high-level features;

[0070] Step S4: feature fusion and classification, adjusting the number of output nodes of the FC layer in the improved CNN network and optimizing the classification of the improved CNN network;

[0071] Step S5: training and testing, dividing the preprocessed data into a training set and a test set in a ratio of 4:1, using the training set to train the improved CNN network, using the test set data for verification, and using the cross entropy loss function and the triplet loss function to evaluate and optimize the improved CNN network;

[0072] Step S6: Hyperparameter tuning, using the Bayesian optimization method to optimize the hyperparameters of the improved CNN network to obtain the final temporal relative attention CNN network model;

[0073] Step S7: Network model application, using the final time relative attention CNN network model to build a power grid operation status evaluation model, evaluate the power grid operation status, and obtain evaluation data.

[0074] Embodiment 5, based on the above embodiment, in step S2, the backbone network design specifically includes the following steps:

[0075] Step S21: Use 7×7 convolution and maximum pooling method to perform downsampling operation to extract low-level features;

[0076] Step S22: Use the average pooling method to reduce the dimension of low-level features, and use the Dropout operation to randomly discard part of the output to extract the overall features;

[0077] Step S23: Use the time series relationship layer to adjust and capture the dynamic changes of power grid data;

[0078] Step S24: self-attention mechanism, using DCSA to dynamically generate convolution kernels, optimizes the learning ability of the CNN network in feature extraction, effectively limits the consumption of calculations and network parameters, and improves the flexibility and efficiency of the CNN network;

[0079] Step S25: Channel and spatial attention mechanism, using the DANet mechanism to allocate attention to the channel dimension and spatial dimension of the CNN network, optimize the feature extraction results, and enhance the CNN network's ability to understand complex scenes;

[0080] Step S26: Normalization and activation, use the GN layer to divide the input preprocessed data into M groups and calculate the mean and variance of each group for normalization. At the same time, use the ReLU activation function to optimize the CNN network.

[0081] Through the above operations, the existing monitoring system still relies on traditional statistical analysis methods and lacks support for big data and advanced algorithms, which makes it difficult for the system to deeply explore the value behind the data. The time relative attention CNN network model is used to analyze the safety of the power grid. According to the changes in the time series, the operation status of the power grid is efficiently extracted and pattern recognized. By introducing the consideration of the time dimension, the model can more keenly capture the characteristics of dynamic changes in the operation of the power grid, and realize comprehensive and accurate monitoring of the operation status of the power grid;

[0082] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The fault diagnosis and early warning module uses a method based on a U-Net fast global self-attention network to build a fault diagnosis model. The method based on a U-Net fast global self-attention network specifically includes the following steps:

[0083] Step M1: construct an encoder and use the U-Net network encoder to extract features from the evaluation data of the fault diagnosis and early warning module to obtain reduced-dimensional feature data;

[0084] Step M2: Fast global self-attention mechanism, using sum normalization to calculate the attention weight, using cyclic multiplication and FFT to calculate the attention matrix, and constructing a fast global self-attention mechanism;

[0085] Step M3: Skip connection: Skip connections are set between different layers of the U-Net network encoder to obtain feature information of different scales.

[0086] Step M4: Construct a decoder, use the U-Net network decoder to process feature information of different scales, and output fault identification results;

[0087] Step M5: Loss function optimization, using binary cross entropy loss and optimization algorithm to train and optimize the U-Net-based fast global self-attention network model to obtain the final fault diagnosis model.

[0088] Embodiment 7, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, step M1, constructing an encoder, specifically comprising the following steps:

[0089] Step M11: Patch embedding, using a 2×2×2 convolutional layer to perform patch embedding on the evaluation data, converting it into a low-resolution feature space, and outputting embedded feature data;

[0090] Step M12: Transformer self-attention module, using N Transformer self-attention modules to extract deep features from the embedded feature data, wherein the Transformer self-attention module includes layer normalization, a fast global self-attention module, a residual connection, and a multi-layer perceptron;

[0091] Step M13: Patch merging. After the first Transformer self-attention module, the adjacent 2×2×2 convolutional layers are connected using a patch merging layer and processed using a linear layer to obtain reduced-dimensional feature data.

[0092] Embodiment 8, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and in step M2, specifically includes the following steps:

[0093] Step M21: Use sum normalization and cyclic multiplication to construct fast global self-attention. The formula used is as follows:

[0094]

[0095] Where Q is the query matrix, K is the key matrix, V is the value matrix, and D K represents the dimension of the key matrix K, Q 2 and K 2 represent the squares of the elements in Q and K respectively, is the transposed symbol, Z is a matrix of all 1s, Represents a cyclic multiplication operation, F SA (Q, K, V) represents the features of the fast global self-attention output;

[0096] Step M22: Circular multiplication calculation, the formula used is as follows:

[0097]

[0098] Where IFFT() denotes inverse fast Fourier transform, FFT() denotes fast Fourier transform, i and j denote the column and row indices of the matrix respectively, C denotes the channel dimension, and * denotes element-wise multiplication.

[0099] Embodiment 9, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step M4, a decoder is constructed, which specifically includes the following steps:

[0100] Step M41: feature mapping, using the SERes block to perform feature mapping on the feature information of different scales transmitted by the skip connection to obtain the mapped feature data;

[0101] Step M42: feature upsampling, upsampling the mapped feature data using 3D interpolation to restore it to the size of the original evaluation data, thereby obtaining upsampled feature data;

[0102] Step M43: feature concatenation, concatenating the upsampled feature data with the embedded feature data to form a complete feature map;

[0103] Step M44: Output the result, use a Sigmoid layer to calculate the final fault recognition probability, and output the recognition result.

[0104] Through the above operations, in view of the limited performance of existing fault diagnosis models in identifying and locating complex faults, and the high computational cost and memory consumption, a fault diagnosis model is constructed using a method based on a fast global self-attention network based on U-Net. This method adopts a multi-scale feature extraction strategy, combined with patch embedding, patch merging and skip connection technology, which can capture subtle features in power grid operation data from different scales, effectively integrate local and global information, enhance the recognition accuracy of target areas, and use a fast global self-attention mechanism to reduce the high computational cost and memory overhead required to calculate large attention matrices in traditional self-attention mechanisms, accelerate the calculation process, and greatly reduce memory usage, so that the model can run efficiently in a resource-constrained environment.

[0105] 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.

[0106] 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.

[0107] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A power grid security remote monitoring and management system based on artificial intelligence, characterized by: It includes data acquisition and transmission module, data processing and analysis module, fault diagnosis and early warning module, dispatching and control module, user interaction module and maintenance support module; The data acquisition and transmission module uses sensors and smart meter devices to collect data from the power grid, and transmits the data to the data processing and analysis module through wired and wireless networks to obtain initial data; The data processing and analysis module uses the time relative attention CNN network model to identify changes in initial data, build a power grid operation status assessment model, and generate assessment data; The fault diagnosis and early warning module uses a U-Net-based fast global self-attention network method to build a fault diagnosis model based on the evaluation data, identify the existing fault risk points, and achieve early warning of faults. By setting thresholds, it can send an alarm to the operator before the fault occurs and generate a warning report; The dispatch control module uses machine learning technology to optimize the grid operation strategy, realize the automatic control function, automatically adjust the load distribution and power generation plan, and generate control records; The user interaction module provides a user-friendly graphical interface, displays the data generated by all modules, supports access by multiple terminal devices, and enables users to understand the status of the power grid anytime and anywhere; The maintenance support module has a safety protection mechanism, provides daily operation and maintenance services for the system, and provides effective solutions when faced with unexpected problems.

2. According to claim 1, the artificial intelligence-based power grid security remote monitoring and management system is characterized by: The data processing and analysis module uses the time relative attention CNN network model to identify changes in initial data, specifically including the following steps: Step S1: Data preprocessing, dividing the initial data into continuous sequence data, and preprocessing using one-hot encoding, regularization and standardization methods to obtain preprocessed data; Step S2: backbone network design, using the basic structure of the CNN network to build the network, and using the attention mechanism to optimize the CNN network to obtain an improved CNN network; Step S3: feature extraction, using the improved CNN network to extract features and obtain high-level features; Step S4: feature fusion and classification, adjusting the number of output nodes of the FC layer in the improved CNN network and optimizing the classification of the improved CNN network; Step S5: training and testing, dividing the preprocessed data into a training set and a test set in a ratio of 4:1, using the training set to train the improved CNN network, using the test set data for verification, and using the cross entropy loss function and the triplet loss function to evaluate and optimize the improved CNN network; Step S6: Hyperparameter tuning, using the Bayesian optimization method to optimize the hyperparameters of the improved CNN network to obtain the final temporal relative attention CNN network model; Step S7: Network model application, using the final time relative attention CNN network model to build a power grid operation status evaluation model, evaluate the power grid operation status, and obtain evaluation data.

3. The artificial intelligence-based power grid security remote monitoring and management system according to claim 2 is characterized by: In step S2, the backbone network design specifically includes the following steps: Step S21: Use 7×7 convolution and maximum pooling method to perform downsampling operation to extract low-level features; Step S22: Use the average pooling method to reduce the dimension of low-level features, and use the Dropout operation to randomly discard part of the output to extract the overall features; Step S23: Use the time series relationship layer to adjust and capture the dynamic changes of power grid data; Step S24: self-attention mechanism, using DCSA to dynamically generate convolution kernels to optimize the learning ability of the CNN network in feature extraction; Step S25: channel and spatial attention mechanism, using the DANet mechanism to allocate attention to the channel dimension and spatial dimension of the CNN network to optimize the feature extraction results; Step S26: Normalization and activation, use the GN layer to divide the input preprocessed data into M groups and calculate the mean and variance of each group for normalization. At the same time, use the ReLU activation function to optimize the CNN network.

4. The artificial intelligence-based power grid security remote monitoring and management system according to claim 1 is characterized by: The method of the U-Net-based fast global self-attention network specifically comprises the following steps: Step M1: construct an encoder and use the U-Net network encoder to extract features from the evaluation data of the fault diagnosis and early warning module to obtain reduced-dimensional feature data; Step M2: Fast global self-attention mechanism, using sum normalization to calculate the attention weight, using cyclic multiplication and FFT to calculate the attention matrix, and constructing a fast global self-attention mechanism; Step M3: Skip connection: Skip connections are set between different layers of the U-Net network encoder to obtain feature information of different scales. Step M4: Construct a decoder, use the U-Net network decoder to process feature information of different scales, and output fault identification results; Step M5: Loss function optimization, using binary cross entropy loss and optimization algorithm to train and optimize the U-Net-based fast global self-attention network model to obtain the final fault diagnosis model.

5. The artificial intelligence-based power grid security remote monitoring and management system according to claim 4 is characterized by: In step M2, the following steps are specifically included: Step M21: Use sum normalization and cyclic multiplication to construct fast global self-attention. The formula used is as follows: Where Q is the query matrix, K is the key matrix, V is the value matrix, and D K represents the dimension of the key matrix K, Q 2 and K 2 represent the squares of the elements in Q and K respectively, is the transposed symbol, Z is a matrix of all 1s, Represents a cyclic multiplication operation, F SA (Q, K, V) represents the features of the fast global self-attention output; Step M22: Circular multiplication calculation, the formula used is as follows: Where IFFT() denotes inverse fast Fourier transform, FFT() denotes fast Fourier transform, i and j denote the column and row indices of the matrix respectively, C denotes the channel dimension, and * denotes element-wise multiplication.

Images