Feature fusion-based transmission tower grounding body detection method and related equipment

Abstract

The invention discloses a transmission tower grounding body detection method based on feature fusion and related equipment, and relates to the technical field of power equipment, and the method comprises the steps: carrying out the multi-dimensional data feature extraction of an obtained three-axis orthogonal coil electromagnetic signal and an environment state parameter through a cyclic convolutional neural network; performing multi-dimensional data feature fusion analysis by adopting a cascade echo state network, and constructing a grounding body detection model to output a burial depth detection result of the grounding body; features are extracted through a one-dimensional cyclic convolutional network, an accurate and comprehensive data basis is provided for detection, fusion analysis is carried out on the features, the calculation amount is reduced, the calculation efficiency is improved, and the defect that a conventional grounding body detection method is difficult to consider both the calculation efficiency and the calculation precision due to large analysis errors of single-coil signals and unstable data quality is overcome. The grounding body burial depth detection device and the grounding body burial depth detection method have the advantages that the detection efficiency is improved, and meanwhile, the detection accuracy is considered.

Description

Technical Field

[0001] This application relates to the technical field of power equipment, and specifically to a detection method and related equipment for transmission tower grounding bodies based on feature fusion. Background Art

[0002] In the current engineering application field, accurately detecting the burial depth of the grounding device is a crucial link, which is directly related to the safe and stable operation of the power system. In conventional technical means, this task relies on a single coil to receive the electromagnetic signals emitted by the grounding device. However, this method has obvious limitations: it can only capture the electromagnetic information of a single plane, which not only limits the comprehensiveness of signal reception but also may lead to signal reception failure or large signal errors; due to these limitations, the conventional single-coil measurement method often fails to achieve the expected accuracy and reliability in practical applications.

[0003] With the rapid development of artificial intelligence technology, intelligent detection methods have shown great potential in various fields. However, when dealing with complex and diverse information, the data processing ability and prediction effect of conventional artificial intelligence methods often fail to meet the requirements of practical applications. In particular, both data quality and algorithm complexity affect the calculation accuracy and calculation efficiency. For example, according to factors such as the height of the grounding body, wire tension, and wind load, combined with the stability requirements of the pole, the demand for the burial depth of the pole is predicted. However, this method needs to comprehensively consider various factors, making it difficult to ensure the reliability of data quality, and thus resulting in unreliable prediction results.

[0004] The patent "A Detection Method and System for the Horizontal Topology Structure of a Substation Grounding Grid", publication number: CN115421199A, publication date: December 2, 2022, discloses that by applying a heterodyne excitation signal to the grounding grid, then detecting the grounding grid to which the heterodyne excitation signal is applied, taking the magnetic induction intensity data perpendicular to the direction of the grounding body, and calculating the difference between adjacent data respectively according to row arrangement and column arrangement; according to the difference calculation, reconstructing the differential array of the magnetic induction intensity data; performing differential calculation on the reconstructed differential array of the magnetic induction intensity data to obtain the position and topology structure of the grounding grid conductor; on the basis of applying the heterodyne excitation signal, through difference calculation and differential calculation, the horizontal topology structure of the substation grounding grid can be quickly drawn. However, this solution only obtains the magnetic induction intensity data perpendicular to the direction of the grounding body, with single electromagnetic information, and cannot accurately evaluate the magnetic induction intensity in the three-dimensional orthogonal directions of the grounding body, thus unable to achieve accurate and reliable detection of the burial depth of the grounding body. Summary of the Invention

[0005] This application addresses the problem that conventional grounding body detection methods have a single signal source and cannot balance computational efficiency and computational accuracy, resulting in unreliable grounding body depth detection. A transmission tower grounding body detection method and related equipment based on feature fusion are proposed. Multi-dimensional data feature extraction is performed on the acquired three-axis orthogonal coil electromagnetic signals and environmental state parameters through a recurrent convolutional neural network, and a cascade echo state network is used to perform multi-dimensional data feature fusion analysis, and a grounding body detection model is constructed to output the grounding body depth detection result. Feature extraction through a one-dimensional recurrent convolutional network provides an accurate and comprehensive data basis for detection, and feature fusion analysis reduces the amount of calculation and improves computational efficiency. This overcomes the problem that conventional grounding body detection methods have large single-coil signal analysis errors and unstable data quality, making it difficult to balance computational efficiency and computational accuracy, resulting in unreliable grounding body depth detection. This achieves the goal of improving detection efficiency while taking into account detection accuracy.

[0006] To solve the above technical problems, the technical solutions adopted in the embodiments of the present application are as follows: In a first aspect, embodiments of the present application provide a method and related equipment for detecting a grounding body of a transmission tower based on feature fusion, comprising the following steps: Acquire electromagnetic signal data of the three-axis orthogonal coil and environmental state parameters of the ground body, and initialize the electromagnetic signal data and the environmental state parameters into a target data sequence; Performing feature extraction on the target data sequence based on a recurrent convolutional neural network to obtain multidimensional data features; fusing the multi-dimensional data features according to a cascade echo state network to construct a ground body detection model; The multi-dimensional data features are detected based on the grounding body detection model, and the grounding body burial depth detection result is output.

[0007] Preferably, the acquiring of electromagnetic signal data of the three-axis orthogonal coil and environmental state parameters of the grounding body comprises: collecting three-dimensional electromagnetic signal data based on the three-axis orthogonal coil layout in response to a transmission current introduced by the grounding device; Synchronously, the soil resistivity sensor, temperature sensor and humidity sensor respectively collect three-axis soil moisture, temperature and conductivity to obtain the environmental state parameters of the grounding body.

[0008] Preferably, the initializing the electromagnetic signal data and the environmental state parameters into a target data sequence comprises: sorting the electromagnetic signal data and the environmental state parameters based on time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; Perform outlier removal, missing value filling, denoising, and standardization on the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain a number of signal data sequences and a number of environmental state parameter sequences; Concatenate the signal data sequence and the environmental state parameter sequence to obtain the target data sequence.

[0009] Preferably, the feature extraction of the signal data sequence based on the recurrent convolutional neural network to obtain multi-dimensional data features includes: Segment the target data sequence of each axis according to the time step to obtain a data sliding window of a fixed length; Perform global data feature extraction based on the data sliding window; Normalize the global data features based on the residual block, and perform dimensionality reduction processing on the global data features through a fully connected layer to obtain multi-dimensional abstract features of the three-axis orthogonal coil, including at least the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend.

[0010] Preferably, the performing global data feature extraction based on the data sliding window includes: Perform parallel convolution operations on the target data sequence according to a number of one-dimensional convolution kernels, and output local data features including at least the time-domain fluctuation of the electromagnetic signal, the frequency-domain energy distribution, and the environmental noise state; Aggregate the local data features in the time dimension through a pooling layer to obtain the global feature of the electromagnetic signal and the global feature of the environmental parameters.

[0011] Preferably, the fusing the multi-dimensional data features according to the cascaded echo state network to construct a grounding body detection model includes: Perform feature splicing on the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend to obtain a target feature vector; Initialize the input weight matrix, the recurrent weight matrix, the leakage rate, the neuron activation vector, and the update state based on the dimension of the target feature vector; Update the state parameters of each layer cascaded reserve pool sequentially according to the time step, and merge the states of all cascaded reserve pools at the last time step to establish a state matrix; Construct a grounding body prediction model based on the state matrix.

[0012] Preferably, the base sequentially updates the state parameters of each hierarchical cascaded reservoir according to the time step, including: adjusting the initial values of the input weights and recurrent weights of each hierarchical cascaded reservoir according to the mean and variance of the target feature vector, and dynamically adjusting the leakage rate and spectral radius according to the input weights and recurrent weights for the absolute eigenvalues.

[0013] In a second aspect, an embodiment of the present application provides a transmission tower grounding body detection device based on feature fusion, including: a data acquisition module, configured to acquire electromagnetic signal data of a three-axis orthogonal coil and environmental state parameters of the grounding body, and initialize the electromagnetic signal data and the environmental state parameters into a target data sequence; a feature extraction module, configured to extract features from the target data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features; a feature analysis module, configured to fuse the multi-dimensional data features according to a cascaded echo state network to construct a grounding body detection model.

[0014] In a third aspect, an embodiment of the present application provides a computer device, including: a processor, a memory, and a bus, where the memory stores machine-readable instructions executable by the processor. When the computer device runs, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of the transmission tower grounding body detection method based on feature fusion as described in the first aspect above.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it performs the steps of the transmission tower grounding body detection method based on feature fusion as described in the first aspect above.

[0016] Advantages of the present application: By using a three-axis orthogonal coil that is pairwise orthogonal to obtain electromagnetic signals, the defect that a single coil cannot sense weak electromagnetic induction signals because it is not perpendicular to the magnetic induction line can be solved. Combining environmental parameters as input features fully considers the influence of environmental factors such as soil humidity, temperature, and conductivity on the propagation of electromagnetic signals, improves the robustness of feature fusion analysis in complex environments, and can reduce the false detection rate; By using multi-scale convolutional kernels of a recurrent convolutional neural network to perform parallel processing on the target data sequence, data features can be captured more precisely from different scales to reduce the subsequent data calculation amount; Through the initialization strategy of feature statistical characteristics, the initial value of the weight is adjusted according to the mean and variance of the target feature vector, enabling the reservoir to adapt to the input data features faster. Meanwhile, by dynamically adjusting the leakage rate and spectral radius, the reservoir can quickly explore the state space and more stably learn the long-term dependence relationship of data features, achieving accurate prediction of the grounding electrode burial depth. Brief Description of the Drawings

[0017] Other features, objectives, and advantages of the present application will become more apparent by reading the detailed description of the non-restrictive embodiments with reference to the following drawings. The drawings are only for the purpose of showing the preferred embodiments and are not considered as a limitation to the present application. Moreover, throughout the drawings, the same reference numerals are used to represent the same components.

[0018] Figure 1 Flowchart of a method for detecting grounding electrodes of transmission towers based on feature fusion provided by an embodiment of the present application.

[0019] Figure 2 Schematic diagram of a three-axis orthogonal coil structure provided by an embodiment of the present application.

[0020] Figure 3 Schematic diagram of the module of a device for detecting grounding electrodes of transmission towers based on feature fusion provided by an embodiment of the present application.

[0021] Figure 4 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Detailed Embodiments

[0022] To make the objectives, technical solutions, and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only the best embodiments of the present application and are only used to explain the present application, without limiting the protection scope of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present application.

[0023] Embodiment 1: As Figure 1 shown, a method for detecting grounding electrodes of transmission towers based on feature fusion includes the following steps: S101. Obtain the electromagnetic signal data of the three-axis orthogonal coil and the environmental state parameters of the grounding electrode, and initialize the electromagnetic signal data and the environmental state parameters as a target data sequence.

[0024] Specifically, the obtaining of the electromagnetic signal data of the three-axis orthogonal coil and the environmental state parameters of the grounding electrode includes: in response to the transmitted current introduced by the grounding device, collecting three-dimensional electromagnetic signal data based on the layout of the three-axis orthogonal coil; Synchronously, according to the soil resistivity sensor, temperature sensor, and humidity sensor, the soil humidity, temperature, and conductivity of the three axes are respectively collected to obtain the environmental state parameters of the grounding electrode.

[0025] Specifically, the initializing the electromagnetic signal data and the environmental state parameters into a target data sequence includes: respectively organizing the electromagnetic signal data and the environmental state parameters based on the time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; Performing outlier removal, missing value filling, denoising, and standardization on the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain a plurality of signal data sequences and a plurality of environmental state parameter sequences; Splicing the signal data sequences and the environmental state parameter sequences to obtain the target data sequence.

[0026] It can be understood that when using single-coil magnetic induction measurement, the receiving coil receives the magnetic induction intensity emitted by the grounding device. According to Faraday's law of electromagnetic induction, the induced electromotive force of the receiving coil is only related to the position of the receiving coil and the angle of the three-axis plane component of the magnetic induction intensity at the center of the coil. Therefore, the angle of the receiving coil has a large interference on the induced electromotive force, and once the electromagnetic signal does not just pass through the center point of the coil, the electromagnetic signal cannot be received, and there is a large error in the received electromagnetic signal. When measuring the grounding device, the position of the grounding device cannot be clearly located, so the direction of the magnetic induction line cannot be known. Especially for grounding devices buried more than 1.5 meters deep, their magnetic induction intensity is very weak. If the receiving coil cannot be parallel to the magnetic induction line it emits, the existence of the grounding resistance here cannot be detected. Therefore, the present application performs induction measurement through a three-axis orthogonal coil.

[0027] Furthermore, as Figure 2 shown, when using a three-axis orthogonal coil with pairwise orthogonality, the center points of the three coils coincide, and the coil radii and turns are also the same. Among them, λ is in the same direction as the magnetic induction intensity, and λ1, λ2, and λ3 are the normal vectors of the three-axis orthogonal coil respectively.

[0028] When the grounding device conducts the transmitting current, the induced electromotive forces of the three coils at a certain point on the ground of the three-axis orthogonal coil are expressed as follows: U i =-B p I0ωcos(ωT)[acosα i +bcosβ i +ccosγ i (1) In the formula, i = 1, 2, 3 represents the coil number; a = 3xz, b = 3yz, c = 2z 2 -x 2 -y2 , where x, y, and z are the coil center coordinates; α i , β i , γ i respectively represent the angles between the centers of the three coils and the coordinate axes; B is the magnetic induction intensity; T represents the period, and ω represents the angular frequency; Therefore, the induced electromotive force of the receiving coil is expressed as follows: In the formula, ψ = -B p I0ωcos(ωT); It can be seen that the induced electromotive force U(T) of the three-axis orthogonal coil has nothing to do with the rotation angle of the coil. No matter how the coil rotates, the induced electromotive force at this point is a constant value. Therefore, the three-axis coil can solve the defect that a single coil cannot detect weak electromagnetic induction signals because it is not perpendicular to the magnetic induction line.

[0029] In this embodiment, three orthogonal coils are used to synchronously collect three-dimensional electromagnetic signals, which can not only provide multi-dimensional raw data, but also eliminate signal loss or errors caused by angle deviation of a single coil, ensuring that the angle between the coil normal vector and the magnetic induction line does not affect the induced electromotive force; considering that factors such as soil humidity and temperature may affect the propagation of electromagnetic signals, the data of sensors such as resistivity, temperature, and humidity are collected to compensate for the influence of electromagnetic signal propagation, providing strong data support for effectively detecting the buried depth value of the grounding body.

[0030] S102. Extract features from the target data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features.

[0031] Specifically, S102 includes: Arrange the electromagnetic signal data and the environmental state parameters respectively based on the time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; Remove outliers, fill in missing values, denoise, and standardize the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain several signal data sequences and several environmental state parameter sequences; Concatenate the signal data sequences and the environmental state parameter sequences to obtain the target data sequence.

[0032] In this embodiment, outlier, missing value, denoising, and normalization processing are performed on the electromagnetic signal data. The purpose is to filter out abnormally fluctuating electromagnetic signals, remove the noise components in the data to retain the effective electromagnetic signal characteristics, and avoid data quality problems caused by measurement errors or device grounding device failures; outlier, missing value, denoising, and normalization processing are performed on the environmental state parameters. The purpose is to remove the parameters under extreme climates and the abnormal data caused by sensor failures; provide accurate data support for subsequent feature extraction and fusion analysis. Finally, the electromagnetic signal data and environmental state parameters are standardized into a data format compatible with the recurrent convolutional neural network, reducing the data compatibility processing flow in the model, so as to improve the feature extraction efficiency and further improve the grounding body burial depth detection efficiency.

[0033] Specifically, the feature extraction of the signal data sequence based on the recurrent convolutional neural network to obtain multi-dimensional data features includes: The target data sequence of each axis is segmented according to the time step to obtain a data sliding window with a fixed length; Global data feature extraction is performed based on the data sliding window; Based on the residual block, the global data features are normalized, and the global data features are dimensionally reduced through a fully connected layer to obtain multi-dimensional abstract features of the three-axis orthogonal coil, including at least the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend.

[0034] Specifically, the global data feature extraction based on the data sliding window includes: Parallel convolution operations are performed on the target data sequence according to several one-dimensional convolution kernels, and local data features including at least the time-domain fluctuation of the electromagnetic signal, the frequency-domain energy distribution, and the environmental noise state are output; The local data features are aggregated in the time dimension through a pooling layer to obtain the global features of the electromagnetic signal and the global features of the environmental parameters.

[0035] Optionally, when extracting local features through a convolutional layer, pre-normalization processing is performed on the electromagnetic signal vector and the environmental parameter vector at time t. For example, the minimum-maximum normalization method is used. The normalized electromagnetic signal and environmental parameters are concatenated into a target data sequence, and a one-dimensional convolution kernel is used to perform a convolution operation on the target data sequence. Let the convolution kernel be the data sliding window.

[0036] In some embodiments, a one-dimensional deep residual model applicable to time series signal analysis can be designed by establishing a cross-layer mapping path and organically integrating optimization strategies such as feature dimensionality reduction pooling, batch distribution regularization, and neuron dropout. This model undertakes the dual functions of primary feature encoding and secondary feature decoding in the hierarchical feature fusion system. The structure of the residual block includes two convolutional layers, two batch normalization layers, and a key residual connection. This design enhances the network performance by introducing the mechanism of skip connections. In this architecture, each convolutional layer is responsible for feature extraction, while the batch normalization layer is used to accelerate the training process and improve the model stability. As a special connection method, the residual connection allows the input to directly skip one or more layers and be added to the output of the subsequent layers on the forward propagation path, effectively alleviating the vanishing gradient problem in deep neural networks and thus promoting the learning ability of deep networks.

[0037] Furthermore, during the network training phase, the dropout technique is implemented in the fully connected layer. By randomly and temporarily "turning off" some neurons, it aims to weaken the intricate co-adaptability between the neuron nodes in the fully connected layer, thereby enhancing the generalization performance of the one-dimensional recurrent convolutional neural network.

[0038] Specifically, the target data sequences of each axis are input into the one-dimensional recurrent convolutional neural network. The network starts with a series connection of a convolutional layer and a pooling layer, which is responsible for initially extracting the global features of the input data. A series of residual blocks and pooling layers are alternately stacked to perform deep feature extraction tasks to further capture the potential information in the data. Finally, the target data sequences pass through two fully connected layers, with a dropout layer interspersed between them to reduce the risk of overfitting, and are transformed into classification probability outputs through an activation function layer, effectively extracting multi-dimensional features from the electromagnetic signals and environmental parameters.

[0039] In this embodiment, the one-dimensional recurrent convolutional neural network independently extracts local features such as the time-domain fluctuations, frequency components, and mutation points of the electromagnetic signal, as well as deep abstract features, from the magnetic induction measurement data obtained from the three-axis orthogonal coils in parallel and the environmental state parameters obtained from each sensor. It can fully capture the non-linear features of the electromagnetic signal, such as power frequency interference and attenuation trends, providing comprehensive data support for subsequent detection.

[0040] S103. Fuse the multi-dimensional data features according to the cascaded echo state network to construct a grounding body detection model.

[0041] Specifically, S103 includes: Perform feature splicing on the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend to obtain a target feature vector; Initialize the input weight matrix, recurrent weight matrix, leakage rate, neuron activation vector, and update status based on the dimension of the target feature vector; Update the state parameters of each hierarchical cascade reservoir sequentially according to the time step, and merge the states of all cascade reservoirs at the last time step to establish a state matrix; Construct a grounding body prediction model based on the state matrix.

[0042] Specifically, updating the state parameters of each hierarchical cascade reservoir sequentially according to the time step includes: adjusting the initial values of the input weights and recurrent weights of each hierarchical cascade reservoir according to the mean and variance of the target feature vector, and dynamically adjusting the leakage rate and spectral radius according to the input weights and recurrent weights.

[0043] S104. Detect the multi-dimensional data features based on the grounding body detection model, and output the detection result of the grounding body burial depth. Among them, predict the target feature vector through the grounding body prediction model, and output the predicted value of the grounding body burial depth.

[0044] As an implementation, for the average intensity E of the electromagnetic signal avg , fluctuation period T wave , main frequency f main , resistivity variance temperature trend T trend and humidity trend H trend adopt the min-max normalization method for processing to ensure that different features are in the same scale range; splice the normalized features into a target feature vector, denoted as Set the reservoir layer of the cascaded echo state network (CasESN) according to the business requirements. According to the dimension of the target feature vector, that is, the type of feature, initialize the input weight matrix and the recurrent weight matrix respectively, and adjust the spectral radius of the recurrent weight matrix; at the same time, initialize the leakage rate and the initial state of each reservoir layer to 0.

[0045] Furthermore, update the state of each reservoir layer from the first layer to the last layer in turn through the mean and variance of the target feature vector, collect the states of all reservoir layers at the last time step to form a state matrix; solve the optimal output weight matrix by minimizing the loss function according to the state matrix to obtain the grounding body prediction model; predict the target feature vector through this model and output the predicted value of the grounding body burial depth.

[0046] Furthermore, when solving for the optimal output weight matrix by minimizing the loss function, the optimal weights can be verified with different regularization parameters. First, select a suitable range of values for the regularization parameter. Use the average intensity, fluctuation period, main frequency, resistivity variance, temperature trend, and humidity trend of the electromagnetic signals in the validation set to update the reservoir state, obtain the state matrix of the validation set, and use the solved output weights for prediction. Calculate the evaluation metrics on the validation set, such as the mean squared error (MSE), mean absolute error (MAE), etc. Record the regularization parameter that optimizes the evaluation metrics of the validation set, and use the optimal regularization parameter to retrain the model on the entire training set to solve for the optimal output weight matrix. The range of values for the regularization parameter can be adjusted according to the actual situation to ensure that a suitable optimal value can be found.

[0047] Among them, CasESN subdivides the original reservoir structure into multiple levels, which are connected to each other in a cascaded manner. The input weights and feedback weights of each sub-reservoir are randomly initialized and remain constant during the training process; in the CasESN model, the spectral radius is defined as the largest absolute eigenvalue of the weight matrix W. For the case of zero input, keeping the spectral radius less than 1 helps to realize the input of data features of the average intensity, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend of the electromagnetic signal at different time scales; for the case of non-zero input, appropriately increasing the value of the spectral radius can optimize the performance of the sub-reservoir, ensure that the characteristics of various influencing factors that affect the propagation of electromagnetic signals in the detection of the burial depth of the grounding electrode can be fully incorporated into the input, and thus improve the prediction accuracy of the network structure.

[0048] In this embodiment, multi-time scale features are captured through cascaded reservoirs to enhance the adaptability to complex electromagnetic signals and environmental state parameters; the cascaded echo state network improves the long-term dependence modeling ability and can reduce the prediction delay; through the cascaded echo state network, efficient fusion of data features is achieved. The connection pattern between neurons in each level of sub-reservoir is characterized by a specific random matrix, and the output of the upper-level sub-reservoir is used as the input of the lower level to realize the hierarchical transmission and processing of information, and thus comprehensive analysis of multi-dimensional features is realized, so as to more comprehensively extract the feature information in the magnetic induction measurement signal. Through feature fusion, multi-dimensional information is effectively integrated, information loss and errors are reduced, and the burial depth of the grounding body of the transmission tower can be detected more accurately.

[0049] In addition, the feature fusion method enables the model to better adapt to complex electromagnetic environments and grounding body conditions. Since multiple dimensions of features are considered simultaneously, the model has stronger adaptability to changes in factors such as different geological conditions, grounding body shapes and positions, etc., improving the reliability and stability of the detection results.

[0050] Based on the same inventive concept, an apparatus 300 for detecting a grounding electrode of a transmission tower based on feature fusion corresponding to the method for detecting a grounding electrode of a transmission tower based on feature fusion is further provided in an embodiment of the present application. Since the principle of solving problems by the apparatus in the embodiment of the present application is similar to that of the above-mentioned method for detecting a grounding electrode of a transmission tower based on feature fusion in the embodiment of the present application, the implementation of the apparatus can refer to the implementation of the method, and the repeated parts will not be described again.

[0051] As Figure 3 shown, the apparatus includes: A data acquisition module 301, configured to acquire electromagnetic signal data of a three-axis orthogonal coil and environmental state parameters of a grounding electrode, and initialize the electromagnetic signal data and the environmental state parameters into a target data sequence; A feature extraction module 302, configured to perform feature extraction on the target data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features; A feature analysis module 303, configured to fuse the multi-dimensional data features according to a cascaded echo state network, construct a grounding electrode detection model to detect the multi-dimensional data features, and output a detection result of the grounding depth of the grounding electrode.

[0052] As an optional implementation manner, the data acquisition module 301 is specifically configured to: In response to the transmitted current introduced by the grounding device, acquire three-dimensional electromagnetic signal data based on the three-axis orthogonal coil layout; Synchronously, acquire the environmental state parameters of the grounding electrode by respectively acquiring the soil humidity, temperature, and conductivity of the three axes according to a soil resistivity sensor, a temperature sensor, and a humidity sensor.

[0053] Arrange the electromagnetic signal data and the environmental state parameters respectively based on the time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; Remove outliers, fill in missing values, denoise, and standardize the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain a plurality of signal data sequences and a plurality of environmental state parameter sequences; Concatenate the signal data sequences and the environmental state parameter sequences to obtain the target data sequence.

[0054] As an optional implementation manner, the feature extraction module 302 is specifically configured to: Sort the electromagnetic signal data and the environmental state parameters respectively based on time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; perform outlier removal, missing value filling, denoising, and standardization on the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain a number of signal data sequences and a number of environmental state parameter sequences; splice the signal data sequences and the environmental state parameter sequences to obtain the target data sequence. Divide the target data sequence of each axis according to the time step to obtain a data sliding window with a fixed length; extract global data features based on the data sliding window; normalize the global data features based on a residual block, and perform dimensionality reduction processing on the global data features through a fully connected layer to obtain multi-dimensional abstract features of the three-axis orthogonal coil, including at least the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend.

[0055] Among them, the extracting global data features based on the data sliding window includes: Perform parallel convolution operations on the target data sequence according to a number of one-dimensional convolution kernels, and output local data features including at least the time-domain fluctuation of the electromagnetic signal, the frequency-domain energy distribution, and the environmental noise state; aggregate the local data features in the time dimension through a pooling layer to obtain the global features of the electromagnetic signal and the global features of the environmental parameters.

[0056] As an optional implementation manner, the feature analysis module 303 is specifically configured to: Perform feature splicing on the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend to obtain a target feature vector; initialize the input weight matrix, the recurrent weight matrix, the leakage rate, the neuron activation vector, and the update state based on the dimension of the target feature vector; sequentially update the state parameters of each hierarchical cascade reservoir according to the time step, and merge the states of all cascade reservoirs at the last time step to establish a state matrix; construct a grounding body prediction model based on the state matrix. Predict the target feature vector through the grounding body prediction model, and output the predicted value of the grounding body burial depth.

[0057] Among them, the sequentially updating the state parameters of each hierarchical cascade reservoir according to the time step includes: Adjust the initial values of the input weights and recurrent weights of each hierarchical cascade reservoir according to the mean and variance of the target feature vector, and dynamically adjust the leakage rate and spectral radius according to the input weights and recurrent weights for the absolute eigenvalue.

[0058] The embodiment of the present application also provides a computer device, such as Figure 4As shown in the figure, it is a schematic structural diagram of a computer device provided by an embodiment of the present application, including: a processor 41, a memory 42, and a bus 43. The memory 42 stores machine-readable instructions executable by the processor 41 (for example, Figure 3 the execution instructions corresponding to the data acquisition module 301, the feature extraction module 302, and the feature analysis module 303 in the device in

[0059] When the computer device runs, communication between the processor 41 and the memory 42 is carried out through the bus 43. When the machine-readable instructions are executed by the processor 41, the steps of the electromagnetic interference modeling method in the above embodiment are executed.

[0060] The above specific implementation manners are the preferred implementation manners of the present application. The specific implementation scope of the present application is not limited thereby. The scope of the present application includes but is not limited to this specific implementation manner. All equivalent changes made according to the shape, structure, and method of the present application are within the protection scope of the present application.

Claims

1. A method for detecting the grounding electrode of a transmission tower based on feature fusion, characterized in that: The method includes the following steps: Obtain the electromagnetic signal data of the three-axis orthogonal coil and the environmental state parameters of the grounding body, and initialize the electromagnetic signal data and the environmental state parameters into a target data sequence; Extract features from the target data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features; Fuse the multi-dimensional data features according to a cascaded echo state network to construct a grounding body detection model; Detect the multi-dimensional data features based on the grounding body detection model and output the detection result of the grounding body burial depth.

2. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 1, wherein: The obtaining of the electromagnetic signal data of the three-axis orthogonal coil and the environmental state parameters of the grounding body includes: In response to the transmitted current introduced by the grounding device, collect three-dimensional electromagnetic signal data based on the layout of the three-axis orthogonal coil; Synchronously, collect the soil humidity, temperature, and conductivity of the three axes according to a soil resistivity sensor, a temperature sensor, and a humidity sensor respectively to obtain the environmental state parameters of the grounding body.

3. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 2, wherein: The initializing of the electromagnetic signal data and the environmental state parameters into a target data sequence includes: Arrange the electromagnetic signal data and the environmental state parameters respectively based on the time series characteristics to obtain an initial signal data sequence and an initial environmental state parameter sequence; Remove outliers, fill in missing values, denoise, and standardize the initial signal data sequence and the initial environmental state parameter sequence respectively to obtain a plurality of signal data sequences and a plurality of environmental state parameter sequences; Concatenate the signal data sequences and the environmental state parameter sequences to obtain the target data sequence.

4. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 1, wherein: The extracting of features from the signal data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features includes: Segment the target data sequence of each axis according to a time step to obtain a data sliding window with a fixed length; Extract global data features based on the data sliding window; Normalize the global data features based on a residual block, and perform dimensionality reduction processing on the global data features through a fully connected layer to obtain multi-dimensional abstract features of the three-axis orthogonal coil, at least including the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend.

5. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 4, wherein: The extracting of global data features based on the data sliding window includes: Perform parallel convolution operations on the target data sequence according to a plurality of one-dimensional convolution kernels, and output local data features at least including the time-domain fluctuation of the electromagnetic signal, the frequency-domain energy distribution, and the environmental noise state; Aggregate the local data features in the time dimension through a pooling layer to obtain the global features of the electromagnetic signal and the global features of the environmental parameters.

6. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 4, characterized in that: The fusing of the multi-dimensional data features according to a cascaded echo state network to construct a grounding body detection model includes: Perform feature concatenation on the average intensity of the electromagnetic signal, the fluctuation period, the main frequency, the resistivity variance, the temperature trend, and the humidity trend to obtain a target feature vector; Initialize the input weight matrix, the recurrent weight matrix, the leakage rate, the neuron activation vector, and the update state based on the dimension of the target feature vector; Update the state parameters of each hierarchical cascaded reservoir sequentially according to the time step, and merge the states of all cascaded reservoirs at the last time step to establish a state matrix; Construct a grounding body prediction model based on the state matrix.

7. The method for detecting the grounding body of a transmission tower based on feature fusion according to claim 6, wherein: The sequential update of the state parameters of each hierarchical cascaded reservoir according to the time step includes: Adjust the initial values of the input weights and recurrent weights of each hierarchical cascaded reservoir according to the mean and variance of the target feature vector, and dynamically adjust the leakage rate and spectral radius of the absolute eigenvalues according to the input weights and recurrent weights.

8. Transmission tower grounding body detection device based on feature fusion, characterized in that: Applicable to the transmission tower grounding body detection method based on feature fusion according to any one of claims 1-7 above, including: A data acquisition module for acquiring electromagnetic signal data of a three-axis orthogonal coil and environmental state parameters of a grounding body, and initializing the electromagnetic signal data and the environmental state parameters into a target data sequence; A feature extraction module for extracting features from the target data sequence based on a recurrent convolutional neural network to obtain multi-dimensional data features; A feature analysis module for fusing the multi-dimensional data features according to a cascaded echo state network to construct a grounding body detection model.

9. A computer device, characterized in that: Including: A processor, a memory, and a bus, where the memory stores machine-readable instructions executable by the processor. When the computer device runs, the processor communicates with the memory through the bus, and the processor executes the machine-readable instructions to perform the steps of the transmission tower grounding body detection method based on feature fusion according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium. When the computer program is run by the processor, it performs the steps of the transmission tower grounding body detection method based on feature fusion according to any one of claims 1 to 7.

Images