Substation grounding grid corrosion detection and positioning method based on improved optimization algorithm

Abstract

The application relates to the field of power system detection, and provides a substation grounding grid corrosion detection and positioning method based on an improved optimization algorithm, a ground penetrating radar device and an electromagnetic induction instrument are used to detect the grounding grid, GPR and EMI data are acquired, the data are input into a MATLAB / Simulink platform for simulation and emulation, time and space convolution processing is respectively performed on the received GPR and EMI data, and time and space dimension feature information is extracted; a self-attention algorithm is applied to the processing result of the time and space convolution, long-range dependence is captured, meanwhile, adaptive gate fusion is used to dynamically allocate weights to different dimension features, and the time and space dimension feature information of the GPR and the EMI is fused; the fused feature information is compared with pre-stored original grounding grid feature information, and a detection result is generated; through the improved time and space convolution neural network and the self-attention mechanism multi-modal data fusion framework, accurate detection and positioning of the substation grounding system corrosion are realized.

Description

Technical Field

[0001] This invention relates to the field of power system detection, specifically a method for detecting and locating corrosion in substation grounding grids based on multimodal data fusion, improved spatiotemporal convolutional neural networks, and self-attention mechanisms, for the precise location of corrosion in substation grounding systems. Background Technology

[0002] With the development of the times, my country's industrial production has entered a new stage. Driven by Industry 4.0, the demand for electricity, as a major energy source in production development, has grown rapidly, bringing new pressures and challenges to the power system. In the power grid, power plants / substations are a crucial link in power transmission, integrating various key equipment such as generators, transformers, secondary equipment, and communication equipment. Many of these are connected to the grounding grid via grounding down conductors. The strong grounding effect and the earth's current dissipation capacity prevent voltage rise in electrical equipment. Therefore, the grounding grid is of great significance in ensuring the safety of power equipment.

[0003] Currently, substation grounding systems are critical infrastructure for the safe operation of power systems, and their integrity and reliability directly affect the stability and security of the entire power grid. However, due to long-term exposure to the complex underground environment, grounding systems inevitably face corrosion problems, which not only reduce the electrical performance of the system but also increase the risk of equipment failure and safety accidents.

[0004] Corrosion in grounding systems is primarily electrochemical corrosion, based on the chemical reaction between metals and ions in an electrolyte environment. When grounding metal materials (such as steel or copper) are exposed to moist soil, they corrode due to oxygen, moisture, and other electrolytes in the soil. Traditional grounding system detection methods, such as visual inspection and resistance measurement, often struggle to identify hidden corrosion problems promptly and accurately, especially in the complex grounding networks of large substations. Furthermore, existing single-mode detection methods, such as ground-penetrating radar (GPR), a physical exploration method that uses ultra-high frequency pulsed electromagnetic waves to detect underground target features, often suffer from insufficient accuracy and weak anti-interference capabilities in complex environments. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for detecting and locating corrosion in substation grounding grids based on an improved optimization algorithm. By using an improved spatiotemporal convolutional neural network (ST-CNN) and a multimodal data fusion framework with a self-attention mechanism, this method is used for the accurate location of corrosion in substation grounding systems, thereby solving the problems in the prior art.

[0006] A method for detecting and locating corrosion in substation grounding grids based on an improved optimization algorithm, comprising:

[0007] Step 1, Grounding grid condition detection: The grounding grid is detected using a ground penetration radar (GPR) device and an electromagnetic induction (EMI) instrument to obtain the GPR and EMI data of the grounding grid;

[0008] Step 2, Multimodal Data Input Stage: Receive GPR and EMI data, and perform simulation using the MATLAB / Simulink platform. The parameter settings of the simulation model refer to the technical specifications and operational data of multimodal data detection and analysis methods applied to substation grounding systems to ensure that the simulation results have practical significance.

[0009] Step 3, Feature Extraction Stage: Spatiotemporal convolution processing is performed on the received GPR and EMI data respectively, and feature information in time and space dimensions is extracted; Feature extraction of GPR and EMI data is performed through an improved spatiotemporal convolutional neural network, with 5 convolutional layers, a kernel size of 3*3, and a hidden state dimension of 128 for the self-attention mechanism.

[0010] Step 4, the self-attention mechanism stage, applies the self-attention algorithm to the processing results of spatiotemporal convolution to capture long-range dependencies and enhance the model's ability to perceive key features. At the same time, it dynamically assigns weights to features of different dimensions through adaptive gating fusion and fuses the temporal and spatial feature information of GPR and EMI.

[0011] Step 5, the output stage, compares the fused feature information with the pre-stored feature information of the original grounding grid under normal conditions to generate the final corrosion detection result.

[0012] Preferably, in step 3, the received GPR and EMI data are subjected to spatiotemporal convolution processing, and temporal and spatial feature information is extracted; specifically, spatial features are extracted from the GPR data, and temporal features are extracted from the EMI data; the specific process for extracting spatial features from the GPR data is as follows:

[0013] The spatial feature extraction of GPR data is performed using a multi-scale convolution algorithm, as shown in the following formula:

[0014]

[0015] Here, conv means extracting features, including spatial features, through convolutional operations of convolutional layers. X represents the input GPR data points, k is the convolutional kernel, and F is the extracted feature map. By changing the size of the convolutional kernel, spatial features at different scales can be captured. i is the scale index for different scales, and k... i For convolution kernels of different scales, This is the feature map of the extracted i-scale GPR data.

[0016] Preferably, the specific process for extracting time features from EMI data is as follows:

[0017] A multi-layer dilated convolutional structure is used to extract temporal features from EMI data. The formula is as follows:

[0018]

[0019] Where l is the number of convolutional layers in the convolutional structure, and lD represents the multi-dilation convolution calculation identifier, meaning that in the very beginning stage of this spatiotemporal convolutional processing, that is, layer 0, no real convolution operation is performed to change the shape of the data.

[0020]

[0021] in, This represents the time-series characteristic map of EMI data at the i-th time scale; ω i express The corresponding fusion weights can be learned through a self-attention mechanism, F EMI This is a feature map of the extracted EMS data.

[0022] Preferably, in step 4, the specific process of applying the self-attention algorithm to the processing result of spatiotemporal convolution is as follows:

[0023] First, the feature map F EMI or F i =F final As input, it is processed through a self-attention mechanism:

[0024] F global =Attention(F final ,F final ,F final )

[0025] This step generates a feature representation F that includes global information. global ;

[0026] Subsequently, the local and global features are fused to obtain the final feature representation:

[0027] F output =Concat(F final ,F global W o

[0028] Among them, W o It is a weight matrix obtained through learning, used to dynamically adjust the importance of local and global features. It can be determined and adjusted through random initialization operations and subsequent training algorithms, including backpropagation.

[0029] Furthermore, to further enhance the model's ability to integrate multi-scale features, a multi-head attention mechanism is introduced. This mechanism learns multiple attention functions in parallel, capturing information from different representation subspaces. Its formula can be expressed as:

[0030] MultiHead(Q,U,V)=Concat(head1,...,head h W o

[0031]

[0032] Where Q, U, and V represent the query vector, key vector, and value vector, respectively, which are derived from the input F. output The vector, obtained through linear transformation, is used to measure the correlation between different elements and extract corresponding value information in the self-attention mechanism. Here, h represents the total number of attention heads, and j is the index of the weight corresponding to the vector. and The learnable weight matrix is ​​a set of parameters used to linearly transform the input data, thereby influencing the computation of query vectors, key vectors, and value vectors, as well as the integration of the final output. They determine which parts of the input data the model focuses on and how it combines information for processing, and can be determined and adjusted through random initialization and subsequent training algorithms, including backpropagation.

[0033] Get the final output head j Then, each head (head1,...,head) h The results of the above operations are concatenated using the Attention operation to obtain a concatenated matrix, which is then combined with the learnable weight matrix. Perform matrix multiplication to obtain the final output feature map F' after processing by the multi-head attention mechanism;

[0034] The F' is evaluated based on the calculation of information entropy and the correlation of tag data. Information entropy reflects the amount of information contained in the feature. For features extracted from GPR and EMI data, the lower the information entropy, the less uncertainty the feature has and the less information is available to effectively distinguish different grounding grid states. Evaluation and screening are carried out according to the preset information entropy threshold.

[0035] The tag data consists of GPR and EMI data under historical grounding grid corrosion conditions. By comparison, the data is evaluated and filtered according to the thresholds set by the staff. The data that passes the evaluation is used as input for subsequent adaptive gating fusion, while the data that fails the evaluation undergoes feature enhancement.

[0036] Preferably, the specific process of dynamically assigning weights to features of different dimensions through adaptive gating fusion is as follows:

[0037] The feature vectors F of GPR and EMI data are processed by improved spatiotemporal convolution and self-attention mechanisms, respectively. GPR 'and F EMI Adaptive gated fusion will calculate a gate vector G to control the fusion ratio of the two features. The gate vector G is calculated as follows:

[0038] G=σ(W g (F GPR :F EMI )+b g )

[0039] Where σ represents the sigmoid activation function, W g and b g These are the learnable weight matrix and the bias term, respectively; ":" indicates the feature concatenation operation.

[0040] Preferably, in step 4, the specific process for fusing the temporal and spatial feature information of GPR and EMI is as follows:

[0041] After obtaining the gating vector, the fusion process of multimodal features can be represented as:

[0042] F fused =G⊙F GPR +(1-G)⊙F EMI

[0043] Where ⊙ represents element-wise multiplication, F fused This refers to the feature information that will be ultimately fused.

[0044] Preferably, in step 5, the fused feature information F fused The feature information of the original grounding grid under normal conditions is compared with the feature information of the pre-stored original grounding grid. The distance metric method, including Euclidean distance or cosine similarity, is used to calculate the degree of difference between the two, and the grounding grid is judged to be corroded and the location of corrosion according to the pre-set threshold.

[0045] Meanwhile, after calculating the corrosion location, the location accuracy is further confirmed. If the location accuracy meets the requirements of subsequent operations, including repair processing, the result is output. If the location accuracy does not meet the requirements of subsequent operations, including repair processing, the parameters of the EMI data after convolution processing are adjusted, including adjusting the convolution kernel size, stride, or padding method, in order to achieve the requirements of accurate spatial feature data.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. This invention utilizes a multimodal data fusion framework based on an improved spatiotemporal convolutional neural network and a self-attention mechanism. By improving the spatiotemporal convolutional neural network's ability to extract spatial features of grounding corrosion (GPR) and temporal features of EMI, it achieves accurate extraction of corrosion information from substation grounding systems. Simultaneously, the introduction of a self-attention mechanism enhances the global correlation between features, improving the accuracy and efficiency of data fusion.

[0048] 2. This invention dynamically integrates multimodal feature technology through an adaptive gating mechanism. During the data fusion stage, this technology employs an adaptive gating mechanism to dynamically integrate data features from both GPR and EMI modes. This mechanism effectively addresses noise and interference in complex environments, improving positioning accuracy and robustness, and providing strong support for the precise monitoring and maintenance of substation grounding networks.

[0049] 3. This invention improves the ability to capture local patterns by modifying the spatiotemporal convolutional neural network, and at the same time uses the self-attention mechanism to model long-range dependencies, thereby achieving a comprehensive and in-depth understanding of the corrosion information of the substation grounding system in the feature extraction stage. This not only improves the utilization rate of data, but also lays a solid foundation for subsequent data fusion and accurate positioning. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the steps in the method of the present invention;

[0051] Figure 2 This is a schematic diagram of the implementation process of the method of the present invention. Detailed Implementation

[0052] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0053] This invention provides a method for detecting and locating corrosion in substation grounding grids based on an improved optimization algorithm, comprising:

[0054] Step 1, Grounding grid condition detection: The grounding grid is detected using a ground penetration radar (GPR) device and an electromagnetic induction (EMI) instrument to obtain the GPR and EMI data of the grounding grid;

[0055] Step 2, Multimodal Data Input Stage: Receive GPR and EMI data, and perform simulation using the MATLAB / Simulink platform. The parameter settings of the simulation model refer to the technical specifications and operational data of multimodal data detection and analysis methods applied to substation grounding systems to ensure that the simulation results have practical significance.

[0056] Step 3, Feature Extraction Stage: Spatiotemporal convolution processing is performed on the received GPR and EMI data respectively, and feature information in time and space dimensions is extracted; Feature extraction of GPR and EMI data is performed through an improved spatiotemporal convolutional neural network, with 5 convolutional layers, a kernel size of 3*3, and a hidden state dimension of 128 for the self-attention mechanism.

[0057] Step 4, the self-attention mechanism stage, applies the self-attention algorithm to the processing results of spatiotemporal convolution to capture long-range dependencies and enhance the model's ability to perceive key features. At the same time, it dynamically assigns weights to features of different dimensions through adaptive gating fusion and fuses the temporal and spatial feature information of GPR and EMI.

[0058] Step 5, the output stage, compares the fused feature information with the pre-stored feature information of the original grounding grid under normal conditions to generate the final corrosion detection result.

[0059] Example:

[0060] like Figure 1-2 As shown in this embodiment, in a large substation, the grounding grid plays a crucial role in ensuring the safe and stable operation of the power system. Due to environmental influences, the grounding grid is easily affected by factors such as electrochemical corrosion, which leads to a decline in grounding performance. Therefore, it is necessary to conduct corrosion detection and location regularly.

[0061] Grounding grid condition detection: Inspectors carry ground penetration radar (GPR) devices and electromagnetic induction (EMI) instruments into the substation.

[0062] Using a grounding detector (GPR), an electromagnetic wave is slowly moved above the grounding grid along a predetermined detection path, emitting electromagnetic waves into the ground. When the electromagnetic waves encounter the grounding grid and different media in and around it, such as normal metal parts, corroded parts, and soil, they are reflected. The GPR receives these reflected waves and records the relevant data.

[0063] Simultaneously, activate the electromagnetic induction (EMI) instrument, which acquires EMI data by sensing changes in the magnetic field around the grounding grid. Throughout the detection process, ensure the detection range covers the entire grounding grid area to obtain comprehensive GPR and EMI data.

[0064] By combining ground-penetrating radar (GPR) with electromagnetic induction (EMI) instruments, multimodal data acquisition was achieved, comprehensively acquiring grounding grid information from different physical mechanisms. This overcomes the problems of insufficient accuracy and weak anti-interference ability faced by traditional single-mode detection methods in complex environments, thereby improving detection accuracy and reliability.

[0065] In the multimodal data input stage, the GPR and EMI data obtained from the substation grounding grid are transmitted to the computer and input into the MATLAB / Simulink platform. Based on the technical specifications for multimodal data detection and analysis methods applied to substation grounding systems and past operational data of the substation's grounding grid, the parameters of the simulation model are set. For example, referring to data such as the substation's soil type, grounding grid material, and past corrosion conditions, the electromagnetic parameters and propagation model parameters in the simulation model are determined to ensure that the simulation results accurately reflect the actual situation of the substation's grounding grid, providing a reliable foundation for subsequent feature extraction and analysis.

[0066] Spatial features are extracted from GPR data using a multi-scale convolution algorithm, as shown in the following formula:

[0067]

[0068] Here, conv means extracting features, including spatial features, through convolutional operations of convolutional layers. X represents the input GPR data points, k is the convolutional kernel, and F is the extracted feature map. By changing the size of the convolutional kernel, spatial features at different scales can be captured. i is the scale index for different scales, and k... i For convolution kernels of different scales, This is the feature map of the extracted i-scale GPR data.

[0069] To extract time-series features from EMI data, a multi-layer dilated convolutional structure is used, as shown in the following formula:

[0070]

[0071] Where l is the number of convolutional layers in the convolutional structure, and lD represents the multi-dilation convolution calculation identifier, meaning that in the very beginning stage of this spatiotemporal convolutional processing, that is, layer 0, no real convolution operation is performed to change the shape of the data.

[0072]

[0073] in, This represents the time-series characteristic map of EMI data at the i-th time scale; ω i express The corresponding fusion weights can be learned through a self-attention mechanism, F EMI This is a feature map of the extracted EMS data.

[0074] In the self-attention mechanism stage, firstly, the feature map F... EMI or F i =F final As input, it is processed through a self-attention mechanism:

[0075] F global =Attention(F final ,F final ,F final )

[0076] This step generates a feature representation F that includes global information. global In this process, the self-attention mechanism calculates the correlation between features, allowing the model to focus on the association between features at different locations. For example, in grounding grid detection, it associates the relationship between corrosion features at different locations to better determine the overall corrosion situation. Subsequently, local and global features are fused to obtain the final feature representation.

[0077] F output =Concat(F final ,F global W o

[0078] Among them, W o It is a weight matrix obtained through learning, used to dynamically adjust the importance of local and global features. It can be determined and adjusted through random initialization operations and subsequent training algorithms, including backpropagation.

[0079] To further enhance the model's ability to integrate multi-scale features, a multi-head attention mechanism is introduced. This mechanism learns multiple attention functions in parallel, capturing information from different representation subspaces. Its formula can be expressed as:

[0080] MultiHead(Q,U,V)=Concat(head1,...,head h W o

[0081]

[0082] Where Q, U, and V represent the query vector, key vector, and value vector, respectively, which are derived from the input F. output The vector, obtained through linear transformation, is used to measure the correlation between different elements and extract corresponding value information in the self-attention mechanism. Here, h represents the total number of attention heads, and j is the index of the weight corresponding to the vector. and The learnable weight matrix is ​​a set of parameters used to linearly transform the input data, thereby influencing the computation of query vectors, key vectors, and value vectors, as well as the integration of the final output. They determine which parts of the input data the model focuses on and how it combines information for processing, and can be determined and adjusted through random initialization and subsequent training algorithms, including backpropagation.

[0083] Get the final output head j Then, each head (head1,...,head) h The results of the above operations are concatenated using the Attention operation to obtain a concatenated matrix, which is then combined with the learnable weight matrix. Perform matrix multiplication to obtain the final output feature map F' after processing by the multi-head attention mechanism;

[0084] The F' is evaluated based on the calculation of information entropy and the correlation of tag data. Information entropy reflects the amount of information contained in the feature. For features extracted from GPR and EMI data, the lower the information entropy, the less uncertainty the feature has and the less information is available to effectively distinguish different grounding grid states. Evaluation and screening are carried out according to the preset information entropy threshold.

[0085] The tag data consists of GPR and EMI data under historical grounding grid corrosion conditions. By comparison, the data is evaluated and filtered according to the thresholds set by the staff. The data that passes the evaluation is used as input for subsequent adaptive gating fusion, while the data that fails the evaluation undergoes feature enhancement.

[0086] For example, for GPR features with low information entropy, the features can be enhanced by increasing the scale range of multi-scale convolution or adjusting the convolution kernel parameters; for EMI features with weak correlation to the label data, the dilation rate or number of layers of dilated convolution can be adjusted to improve the features.

[0087] The specific process of dynamically assigning weights to features of different dimensions through adaptive gating fusion is as follows:

[0088] The feature vectors F of GPR and EMI data are processed by improved spatiotemporal convolution and self-attention mechanisms, respectively. GPR 'and F EMI Adaptive gated fusion will calculate a gate vector G to control the fusion ratio of the two features. The gate vector G is calculated as follows:

[0089] G=σ(W g (F GRP :F EMI )+b g )

[0090] Where σ represents the sigmoid activation function, W g and b g These are the learnable weight matrix and the bias term, respectively; ":" indicates the feature concatenation operation.

[0091] The specific process for fusing the temporal and spatial feature information of GPR and EMI is as follows:

[0092] After obtaining the gating vector, the fusion process of multimodal features can be represented as:

[0093] F fused =G⊙F GPR +(1-G)⊙F EMI

[0094] Where ⊙ represents element-wise multiplication, F fused This refers to the feature information that will be ultimately fused.

[0095] By dynamically integrating multimodal feature technology through an adaptive gating mechanism, this technology dynamically integrates data features from both GPR and EMI modes during the data fusion phase. This mechanism effectively addresses noise and interference in complex environments, improving positioning accuracy and robustness, and providing strong support for the precise monitoring and maintenance of substation grounding networks.

[0096] The fused feature information F fused The feature information of the original grounding grid under normal conditions is compared with the feature information of the pre-stored original grounding grid. The distance metric method, including Euclidean distance or cosine similarity, is used to calculate the degree of difference between the two, and the grounding grid is judged to be corroded and the location of corrosion according to the pre-set threshold.

[0097] Meanwhile, after calculating the corrosion location, the location accuracy is further confirmed. If the location accuracy meets the requirements of subsequent operations, including repair processing, the result is output. If the location accuracy does not meet the requirements of subsequent operations, including repair processing, the parameters of the EMI data after convolution processing are adjusted, including adjusting the convolution kernel size, stride, or padding method, in order to achieve the requirements of accurate spatial feature data.

[0098] In summary, this substation grounding grid corrosion detection and location method based on improved optimization algorithms achieves multimodal data acquisition by integrating ground-penetrating radar (GPR) and electromagnetic induction (EMI) instruments. It also mines and integrates data features using improved spatiotemporal convolutional neural networks, self-attention mechanisms, and multi-head attention mechanisms, and optimizes data fusion through an adaptive gating fusion mechanism. Through multi-dimensional synergy, it comprehensively improves the efficiency and quality of substation grounding grid corrosion detection and location.

[0099] It provides strong technical support for ensuring the safe and stable operation of substations, extending the service life of grounding grids, and promoting the advancement of power system operation and maintenance technology, and has extremely significant economic and social benefits.

[0100] The embodiments of the present invention are given for the purposes of illustration and description. Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for detecting and locating corrosion in substation grounding grids based on an improved optimization algorithm, characterized in that, include: Step 1, Grounding grid condition detection: The grounding grid is detected using a ground penetration radar (GPR) device and an electromagnetic induction (EMI) instrument to obtain the GPR and EMI data of the grounding grid; Step 2, Multimodal Data Input Stage: Receive GPR and EMI data, and perform simulation using the MATLAB / Simulink platform. The parameter settings of the simulation model refer to the technical specifications and operating data of multimodal data detection and analysis methods applied to substation grounding systems to ensure that the simulation results have practical significance. Step 3, Feature Extraction Stage: Spatiotemporal convolution processing is performed on the received GPR and EMI data respectively, and temporal and spatial feature information is extracted. An improved spatiotemporal convolutional neural network is used to extract features from the GPR and EMI data, with 5 convolutional layers and a kernel size of 3.

3. The hidden state dimension of the self-attention mechanism is 128; Step 4, the self-attention mechanism stage, applies the self-attention algorithm to the processing results of spatiotemporal convolution to capture long-range dependencies and enhance the model's ability to perceive key features. At the same time, it dynamically assigns weights to features of different dimensions through adaptive gating fusion and fuses the temporal and spatial feature information of GPR and EMI. Step 5, the output stage, compares the fused feature information with the pre-stored feature information of the original grounding grid under normal conditions to generate the final corrosion detection result.

2. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 1, characterized in that: In step 3, the received GPR and EMI data are subjected to spatiotemporal convolution processing, and temporal and spatial feature information is extracted. Specifically, spatial features are extracted from the GPR data, and temporal features are extracted from the EMI data. The specific process for extracting spatial features from the GPR data is as follows: The spatial feature extraction of GPR data is performed using a multi-scale convolution algorithm, as shown in the following formula: Here, conv means extracting the features that the convolution kernel wants to extract through convolutional operations in convolutional layers, including spatial features. To input GPR data points, For convolution kernel, The extracted feature map is used to capture spatial features at different scales by changing the convolution kernel size, where i is the scale index for different scales. For convolution kernels of different scales, This is the feature map of the extracted i-scale GPR data.

3. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 2, characterized in that: The specific process for extracting time features from EMI data is as follows: A multi-layer dilated convolutional structure is used to extract time-series features from EMI data, as shown in the following formula: in, This represents the number of convolutional layers in the convolutional structure. The identifier for multi-dilation convolution computation means that at the very beginning of this spatiotemporal convolution process, that is, at layer 0, no actual convolution operation was performed to change the shape of the data. in, This represents the time-series characteristic map of EMI data at the i-th time scale; express The corresponding fusion weights can be learned through a self-attention mechanism. This is a feature map of the extracted EMI data.

4. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 1, characterized in that: In step 4, the specific process of applying the self-attention algorithm to the spatiotemporal convolution processing result is as follows: First, the feature map As input, it is processed through a self-attention mechanism: This step generates a feature representation that includes global information. ; Subsequently, the local and global features are fused to obtain the final feature representation: in, It is a weight matrix obtained through learning, used to dynamically adjust the importance of local and global features. It can be determined and adjusted through random initialization operations and subsequent training algorithms, including backpropagation. Furthermore, to further enhance the model's ability to integrate multi-scale features, a multi-head attention mechanism is introduced. This mechanism learns multiple attention functions in parallel, capturing information from different representation subspaces. Its formula can be expressed as: in, , and These represent the query vector, key vector, and value vector, respectively, which are derived from the input. The result, obtained through linear transformation, is used in self-attention mechanisms to measure the degree of correlation between different elements and to extract corresponding value information. For attention head The total number, where j is the index of the vector corresponding to the weight. , and The learnable weight matrix is ​​a set of parameters used to linearly transform the input data, thereby affecting the computation of query vectors, key vectors, and value vectors, as well as the integration of the final output. They determine which parts of the input data the model focuses on and how to combine information for operation. They can be determined and adjusted through random initialization operations and subsequent training algorithms, including backpropagation algorithms. Get the final output Then, each head The results are spliced ​​together. This operation yields a concatenated matrix, which is then compared with the learnable weight matrix. , , Perform matrix multiplication to obtain the final output feature map after multi-head attention mechanism processing. '; Regarding this 'Conduct a quality assessment based on features including information entropy calculation and label data relevance features, and...' 'Calculate information entropy. Information entropy reflects the amount of information contained in a feature. For features extracted from GPR and EMI data, the lower the information entropy, the less uncertainty the feature has and the less information is available to effectively distinguish different grounding grid states. Evaluation and screening are performed based on a preset information entropy threshold.' The tag data consists of GPR and EMI data under historical grounding grid corrosion conditions. By comparison, the data is evaluated and filtered according to the thresholds set by the staff. The data that passes the evaluation is used as input for subsequent adaptive gating fusion, while the data that fails the evaluation undergoes feature enhancement.

5. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 1, characterized in that: The specific process of dynamically assigning weights to features of different dimensions through adaptive gating fusion is as follows: The feature vectors of GPR and EMI data are processed by improved spatiotemporal convolution and self-attention mechanisms, respectively. 'and Adaptive gating fusion will compute a gating vector. The gating vector is used to control the fusion ratio of these two features. The calculation is as follows: in, This represents the sigmoid activation function. and These are the learnable weight matrix and the bias term, respectively, and ":" indicates the feature concatenation operation.

6. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 5, characterized in that: In step 4, the specific process for fusing the temporal and spatial feature information of GPR and EMI is as follows: After obtaining the gating vector, the fusion process of multimodal features can be represented as: in, Indicates element-wise multiplication. This refers to the feature information that will be ultimately fused.

7. The method for detecting and locating corrosion of substation grounding grids based on an improved optimization algorithm as described in claim 6, characterized in that: In step 5, the fused feature information The feature information of the original grounding grid under normal conditions is compared with the feature information of the pre-stored original grounding grid. The distance metric method, including Euclidean distance or cosine similarity, is used to calculate the degree of difference between the two, and the grounding grid is judged to be corroded and the location of corrosion according to the pre-set threshold. Meanwhile, after calculating the corrosion location, the location accuracy is further confirmed. If the location accuracy meets the requirements of subsequent operations, including repair processing, the result is output. If the location accuracy does not meet the requirements of subsequent operations, including repair processing, the parameters of the EMI data after convolution processing are adjusted, including adjusting the convolution kernel size, stride, or padding method, in order to achieve the requirements of accurate spatial feature data.

Images