A comprehensive evaluation method for the performance of filter field grounding system in medium frequency

Abstract

This invention provides a comprehensive evaluation method for the mid-frequency performance of a filter field grounding system, belonging to the field of power grid equipment technology. The invention extracts the resonant frequency distribution by inputting a wideband impedance spectrum into an artificial intelligence frequency domain response identification model based on a coarse spatiotemporal attention mechanism. It then dynamically adjusts the spatial receptive field parameters according to the resonant frequency offset function. A three-dimensional electromagnetic field simulation model is established based on soil frequency domain parameters, and the ground potential and current density distribution are calculated using the radial basis function collocation method and the fast multipole algorithm. The entropy value of current distribution uniformity and the Gini coefficient are calculated to identify current concentration areas and assess local overheating risk. Finally, the analytic hierarchy process (AHP) is used to dynamically weight and comprehensively score indicators such as impedance peak value, resonant frequency, ground potential gradient, current uniformity, and overheating risk. This invention solves the technical problem of inaccurate performance evaluation of filter field grounding systems under mid-frequency impulse conditions.

Description

Technical Field

[0001] This invention belongs to the field of power grid equipment technology, and more specifically, relates to a comprehensive evaluation method for the intermediate frequency performance of a filter field grounding system. Background Technology

[0002] In the grounding safety assessment of power system filter fields, traditional techniques mainly employ power frequency grounding resistance measurement and steady-state potential distribution calculation to evaluate grounding system performance. These methods inject power frequency current into the grounding grid, measure the grounding resistance value, and calculate the surface potential distribution based on a uniform soil model to determine whether the grounding system meets safety requirements. Traditional techniques have good applicability under power frequency conditions and have been widely used in the design and acceptance of grounding systems in substations and power plants. However, traditional techniques have significant drawbacks. First, power frequency measurements cannot reflect the frequency-varying characteristics of the grounding system under medium-frequency impulse conditions, neglecting the frequency dependence of the soil medium and the resonance effect of the grounding grid. Second, the calculation model based on the uniform soil assumption deviates significantly from the actual layered soil structure, leading to inaccurate potential distribution predictions. Third, traditional methods lack quantitative assessment of current distribution uniformity and the risk of local overheating, failing to identify weak points in the grounding system. In existing technologies, the lack of broadband impedance measurement methods for medium-frequency impulse conditions and electromagnetic field simulation models that consider soil frequency-varying characteristics, coupled with the absence of a comprehensive multi-dimensional performance evaluation system, makes it difficult to accurately assess the actual performance of filter field grounding systems under medium-frequency transient disturbances such as lightning strikes or switching overvoltages. In other words, existing technologies suffer from inaccurate performance evaluation of filter field grounding systems under medium-frequency impulse conditions. Summary of the Invention

[0003] In view of this, the present invention provides a comprehensive evaluation method for the mid-frequency performance of a filter field grounding system, which can solve the technical problem of inaccurate performance evaluation of filter field grounding systems under mid-frequency impact conditions in the prior art.

[0004] This invention is implemented as follows: A comprehensive evaluation method for the mid-frequency performance of a filter field grounding system is provided. A wideband impedance measurement device is deployed at the grounding grid nodes to obtain a wideband impedance spectrum curve. Ground-penetrating radar and resistivity imaging technology are used to acquire soil layer structure data and resistivity distribution data to establish a soil frequency variation model and generate a soil frequency domain equivalent parameter dataset. The wideband impedance spectrum curve is input into a frequency domain response identification model, which outputs resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. The spatial receptive field parameters of the coarse spatiotemporal resolution attention mechanism are adjusted based on the resonant frequency offset function value. A three-dimensional electromagnetic field simulation model of the grounding grid is established to obtain the spatial distribution matrix of ground potential and the current density distribution matrix. The entropy value of current distribution uniformity, the Gini coefficient of current concentration, and the local overheating risk index are calculated. The analytic hierarchy process (AHP) is used to calculate the comprehensive weighting coefficient, and the weighted summation of each index yields a comprehensive mid-frequency performance score for the grounding grid. When the comprehensive score is lower than the safety score threshold, weak link location information is generated and output.

[0005] Specifically, the step of obtaining the wideband impedance spectrum curve involves injecting a sweep excitation signal in the range of 0.5kHz to 10kHz into the grounding grid using multi-frequency synchronous injection technology, extracting the response signal through an orthogonal lock-in amplifier, and obtaining the wideband impedance spectrum curve after frequency domain averaging and time domain adaptive filtering.

[0006] Among them, the multi-frequency synchronous injection technology refers to injecting multiple sinusoidal excitation signals of different frequencies into the grounding grid at the same time, and decomposing the response signal into frequency components through Fourier transform to achieve parallel measurement of impedance at multiple frequency points.

[0007] Among them, the quadrature lock-in amplifier is a signal processing device that uses a reference signal and the signal under test to perform correlation operations to extract frequency components, and separates amplitude and phase information through quadrature demodulation technology.

[0008] Among them, frequency domain averaging is a method of reducing the influence of random noise by arithmetically averaging the spectrum data obtained from multiple measurements, while time domain adaptive filtering uses the minimum mean square error criterion to update the filter coefficients.

[0009] Among them, the soil frequency variation model uses fractional differential equations based on Caputo-type fractional derivatives to describe the memory effect of charge diffusion and polarization relaxation in the soil medium, and uses fractional particle swarm optimization algorithm to invert fractional order parameters and relaxation time parameters.

[0010] The frequency domain response identification model includes a feature extraction layer, a coarse spatiotemporal attention layer, and a parameter prediction layer. The feature extraction layer is composed of a one-dimensional convolutional neural network to extract features at multiple scales from the broadband impedance spectrum curve.

[0011] The coarse spatiotemporal attention layer includes a spatial attention module and a temporal attention module. The spatial attention module controls the degree of attention to different frequency bands through spatial receptive field parameters, while the temporal attention module uses a self-attention mechanism to capture long-range dependencies.

[0012] The resonant frequency offset function value is calculated by multiplying the relative deviation between the predicted resonant frequency and the standard resonant frequency, the normalized value of the quality factor, and the normalized value of the impedance by weighting coefficients and then summing them.

[0013] Specifically, when the resonant frequency offset function value is large, the spatial receptive field parameter is increased to expand the range of interest; when the resonant frequency offset function value is small, the spatial receptive field parameter is decreased to focus on local detailed features.

[0014] Among them, the establishment of the three-dimensional electromagnetic field simulation model of the grounding grid adopts the radial basis function collocation method to discretize the surface potential distribution, the boundary element method is used to handle the potential singularity at the grounding conductor boundary, and the fast multipole algorithm is applied to accelerate the calculation in the far field region.

[0015] The calculation of the entropy value of current distribution uniformity involves normalizing the current value of each conductor segment of the grounding grid into a probability distribution, taking the logarithm of the probability, multiplying it by the probability, summing the results, and taking the negative value.

[0016] Specifically, when the entropy value of current distribution uniformity is less than the entropy threshold or the Gini coefficient of current concentration is greater than the Gini coefficient threshold, the current concentration area is marked and the local overheating risk index is calculated.

[0017] The local overheating risk index is calculated by dividing the peak current density by the conductor's allowable current density to obtain the current overload coefficient, dividing the impact duration by the conductor's thermal time constant to obtain the time coefficient, multiplying the two by the heat dissipation correction factor.

[0018] Among them, when the impedance peak value of the broadband impedance spectrum curve is greater than the impedance threshold, or the resonant frequency in the resonant frequency distribution parameter falls into the sensitive frequency band, or the ground potential gradient of the ground potential spatial distribution matrix exceeds the gradient limit, the comprehensive weight coefficient of the corresponding index is increased.

[0019] The performance evaluation report includes basic information about the grounding grid, test results of various performance indicators, comprehensive score of the intermediate frequency performance of the grounding grid, analysis of weak points and improvement suggestions. The weak point location information includes the spatial coordinates of the current concentration area and the location of the measurement point where the ground potential gradient exceeds the gradient limit.

[0020] This invention obtains the frequency domain response characteristics of the grounding grid through broadband impedance measurement. Combined with soil frequency-varying model inversion and three-dimensional electromagnetic field simulation, it establishes a multi-dimensional evaluation index system covering impedance characteristics, resonant behavior, potential distribution, current uniformity, and overheating risk, achieving accurate evaluation of the mid-frequency performance of the filter field grounding system. This invention employs multi-frequency synchronous injection technology and orthogonal lock-in amplifiers to extract the broadband impedance spectrum, overcoming the deficiency of traditional power frequency measurements in reflecting frequency-varying characteristics. It utilizes fractional differential equations to describe the memory effect of soil polarization relaxation and inverts soil frequency domain parameters using layered structure data obtained from ground-penetrating radar and resistivity imaging, solving the simulation bias problem caused by the homogeneous soil assumption. A frequency domain response identification model is introduced to extract the resonant frequency distribution. Current distribution uniformity entropy and Gini coefficient are used to quantify current concentration phenomena, and the weights of each index are dynamically adjusted through the analytic hierarchy process (AHP) to construct a comprehensive scoring system, compensating for the lack of quantitative evaluation in traditional methods. In summary, this invention solves the technical problem of inaccurate performance evaluation of filter field grounding systems under medium-frequency impulse conditions mentioned in the background art by organically combining wideband impedance measurement, soil frequency variation modeling, resonance characteristic identification, and multi-index comprehensive evaluation. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a schematic diagram of the hardware system composition for the comprehensive evaluation of the mid-frequency performance of a filter field grounding system.

[0023] Figure 3 This is a three-dimensional surface diagram of the spatial distribution of ground potential.

[0024] Figure 4 This is a comparison diagram of the current density distribution in the grounding grid.

[0025] Figure 5 Radar chart of scores for each evaluation indicator. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0027] like Figure 1 The diagram shows a flowchart of a comprehensive evaluation method for the mid-frequency performance of a filter field-grounding system provided by this invention. This method includes the following steps: S01. Wideband impedance measurement devices are arranged at multiple nodes of the filter field grounding grid. Multi-frequency synchronous injection technology is used to inject a sweep excitation signal in the range of 0.5kHz to 10kHz into the grounding grid. The response signal is extracted by an orthogonal lock-in amplifier and then processed by frequency domain averaging and time domain adaptive filtering to obtain the wideband impedance spectrum curve. S02. Using ground-penetrating radar and resistivity imaging technology, obtain the layered structure data and resistivity distribution data of the soil around the grounding grid, establish a soil frequency variation model based on fractional differential equations, and use fractional particle swarm optimization algorithm to invert the fractional order parameters and relaxation time parameters of the soil frequency variation model from the broadband impedance spectrum curve obtained in step S01, and generate a soil frequency domain equivalent parameter dataset. S03. Input the broadband impedance spectrum curve obtained in step S01 into the frequency domain response identification model for processing. The frequency domain response identification model outputs resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. Calculate the resonant frequency offset function value and adjust the spatial receptive field parameters of the coarse spatiotemporal resolution attention mechanism of the frequency domain response identification model according to the interval to which the resonant frequency offset function value belongs. S04. Based on the soil frequency domain equivalent parameter dataset generated in step S02, a three-dimensional electromagnetic field simulation model of the grounding grid is established. The radial basis function collocation method is used to discretize and solve the surface potential distribution. The boundary element method is used to handle the potential singularity at the grounding conductor boundary. The fast multipole algorithm is applied to the far field region to accelerate the calculation and obtain the spatial distribution matrix of ground potential and the current density distribution matrix under the medium frequency impulse condition. S05. Calculate the entropy value of current distribution uniformity and the Gini coefficient of current concentration from the current density distribution matrix obtained in step S04. When the entropy value of current distribution uniformity is less than the entropy value threshold or the Gini coefficient of current concentration is greater than the Gini coefficient threshold, mark the current concentration area and calculate the local overheating risk index of the current concentration area. S06. Based on the broadband impedance spectrum curve, resonant frequency distribution parameters, ground potential spatial distribution matrix, current distribution uniformity entropy value, and local overheating risk index obtained in steps S01 to S05, the comprehensive weighting coefficient of each index is calculated using the analytic hierarchy process (AHP). When the impedance peak value of the broadband impedance spectrum curve is greater than the impedance threshold, or the resonant frequency in the resonant frequency distribution parameters falls into the sensitive frequency band, or the ground potential gradient of the ground potential spatial distribution matrix exceeds the gradient limit, the comprehensive weighting coefficient of the corresponding index is increased by 30% to 50%. The weighted summation of each index is used to obtain the comprehensive score value of the intermediate frequency performance of the grounding grid. When the comprehensive score value of the intermediate frequency performance of the grounding grid is lower than the safety score threshold, a performance evaluation report containing weak link location information and improvement suggestions is generated.

[0028] Among them, multi-frequency synchronous injection technology refers to injecting multiple sinusoidal excitation signals of different frequencies into the grounding grid at the same time, and decomposing the response signal into frequency components through Fourier transform to achieve parallel measurement of impedance at multiple frequency points, improving measurement efficiency by 5 to 8 times compared with the traditional point-by-point frequency sweeping method. Quadrature lock-in amplifiers are signal processing devices that extract frequency components by performing correlation operations between a reference signal and the signal under test. Through quadrature demodulation technology, amplitude and phase information are separated, exhibiting strong suppression capabilities against noise and interference, with a signal-to-noise ratio improvement of 40dB to 60dB. Frequency domain averaging is a method of reducing the influence of random noise by arithmetically averaging the spectral data obtained from multiple measurements; the number of measurements is typically 8 to 16. Time-domain adaptive filtering is a dynamic filtering technology that automatically adjusts filter parameters based on signal statistical characteristics. It updates the filter coefficients using the minimum mean square error criterion, and its suppression effect on non-stationary interference is superior to that of fixed-parameter filters.

[0029] Ground-penetrating radar (GPR) is a non-destructive testing device that emits electromagnetic waves and receives reflected waves from the subsurface medium to detect underground structures. Its operating frequency range is 10MHz to 1000MHz, its detection depth is 5m to 30m, and its resolution is 0.1m to 0.5m. Resistivity imaging technology is a geophysical exploration method that uses multiple electrode arrays deployed on the surface to inject current and measure potential, inverting to obtain two-dimensional or three-dimensional images of the subsurface resistivity distribution. The electrode spacing is 1m to 5m, and the imaging depth is 0.3 to 0.5 times the electrode spacing. Layered structure data includes the thickness and depth information of each soil layer. Resistivity distribution data includes the resistivity values ​​and spatial coordinates of each soil layer.

[0030] The soil frequency variation model based on fractional differential equations uses Caputo-type fractional derivatives to describe the memory effect of charge diffusion and polarization relaxation in the soil medium. The fractional order parameter ranges from 0 to 2, reflecting the fractal dimension of the soil's microporous structure. A fractional order parameter closer to 1 indicates a more homogeneous soil structure. The fractional particle swarm optimization algorithm introduces a fractional differential operator into the speed update formula of the standard particle swarm optimization algorithm, enhancing the global search capability and the ability to escape local optima. The convergence performance is optimal when the fractional order parameter is between 0.6 and 0.9. The relaxation time parameter characterizes the timescale of soil polarization establishment and decay, with a value ranging from... s to s is closely related to soil moisture content and ion concentration. The soil frequency domain equivalent parameter dataset includes the fractional-order parameters, the relaxation time parameters, soil resistivity, and soil dielectric constant values ​​at different frequencies.

[0031] The structure of the frequency domain response identification model, considering a coarse spatiotemporal resolution attention mechanism, is as follows: The frequency domain response identification model includes a feature extraction layer, a coarse spatiotemporal attention layer, and a parameter prediction layer. The feature extraction layer consists of four one-dimensional convolutional neural networks with kernel sizes of 16, 32, 64, and 128, a stride of 2, and uses a leaky modified linear unit as the activation function to perform multi-scale feature extraction on the input broadband impedance spectrum curve. The coarse spatiotemporal attention layer includes a spatial attention module and a temporal attention module. The spatial attention module controls the degree of attention to different frequency bands through learnable spatial receptive field parameters, the range of which is specified in the text. The values ​​range from 1 to 20, with larger values ​​indicating a wider frequency range of interest. The time attention module uses a self-attention mechanism to capture the long-range dependence of the broadband impedance spectrum curve at different frequency points. It has 8 attention heads, each with a dimension of 64. The parameter prediction layer consists of 3 fully connected neural networks with 256, 128, and 64 neurons, respectively. It outputs the resonant frequency distribution parameter, the quality factor parameter, and the modal damping parameter. The resonant frequency distribution parameter includes the resonant frequency values ​​and frequency intervals of each order. The quality factor parameter characterizes the sharpness of the resonant peak, and the modal damping parameter reflects the energy dissipation rate. The coarse spatiotemporal resolution attention mechanism achieves coarse-grained control of features in different frequency bands by dynamically adjusting the spatial receptive field parameters. When the resonant frequency offset function value is large, the spatial receptive field parameters are increased to expand the scope of attention and capture coupling characteristics of a wider frequency band. When the resonant frequency offset function value is small, the spatial receptive field parameters are decreased to focus on local detailed features and improve the resonant frequency prediction accuracy. Through this dynamic adjustment, the frequency domain response identification model achieves a balance between computational efficiency and prediction accuracy.

[0032] The steps for establishing the training dataset of the frequency domain response identification model are as follows: 500 sets of broadband impedance spectrum curves under different grounding grid structures and soil conditions are collected as input samples. Each set of input samples contains impedance amplitude and phase data at 200 to 500 frequency points. The true resonant frequency distribution parameters, true quality factor parameters, and true modal damping parameters corresponding to each set of input samples are extracted as label data through impulse response testing and the Prony algorithm. The input samples are normalized, with the impedance amplitude normalized to the 0 to 1 range and the phase normalized to the -1 to 1 range. The training dataset is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0033] The training steps for the frequency domain response identification model are as follows: Mean squared error is used as the loss function; the adaptive moment estimation algorithm is selected as the optimizer; the initial learning rate is set to 0.001, decreasing to 0.5 times the original value every 30 training cycles; the batch size is set to 32, and the number of training cycles is 200; during training, the performance of the frequency domain response identification model is evaluated on the validation set every 10 training cycles; training stops when the validation set loss does not decrease for 20 consecutive training cycles; L2 regularization constraint is applied to the spatial receptive field parameters with a regularization coefficient of 0.0001 to prevent overfitting of the frequency domain response identification model; after training, the resonant frequency prediction error of the frequency domain response identification model is evaluated on the test set, requiring the average relative error to be less than 5%.

[0034] The resonant frequency offset function is used to adjust the spatial receptive field parameters of the coarse spatiotemporal resolution attention mechanism. The resonant frequency offset function is calculated based on the deviation between the predicted resonant frequency and the standard resonant frequency, the quality factor parameter, and the impedance peak value. The resonant frequency offset function is expressed as follows: dividing the difference between the predicted resonant frequency and the standard resonant frequency by the standard resonant frequency yields the relative deviation; dividing the quality factor parameter by the standard quality factor yields the normalized quality factor value; dividing the impedance peak value by the standard impedance peak value yields the normalized impedance value. The relative deviation, ... The normalized quality factor and the normalized impedance are multiplied by weighting coefficients of 0.5, 0.3, and 0.2 respectively, and then summed to obtain the resonant frequency offset function value. When the resonant frequency offset function value is within [0, 0.1), the spatial receptive field parameter is set to 2 to 4; when it is within [0.1, 0.3), the parameter is set to 5 to 10; when it is within [0.3, 0.6), the parameter is set to 11 to 15; and when it is ≥0.6, the parameter is set to 16 to 20. The standard resonant frequency refers to the reference value of the resonant frequency of a typical grounding grid under standard soil conditions, determined by referring to a table based on the grounding grid area and down conductor length. The standard quality factor ranges from 5 to 15, and the standard impedance peak value ranges from 10Ω to 100Ω. The predicted resonant frequency is the resonant frequency value in the resonant frequency distribution parameters output by the frequency domain response identification model. The impedance peak value is the maximum impedance amplitude in the broadband impedance spectrum curve.

[0035] The radial basis function collocation method is a meshless numerical method that uses radial basis functions as interpolation basis functions to select collocation points within the computational domain so that the differential equations are satisfied at these points. The radial basis functions are either multiquadratic functions or Gaussian functions, and the shape parameters are determined through cross-validation. The boundary element method only requires discretization of the boundary and transforms the internal field problem into a boundary integral equation solution using Green's function. It offers better accuracy than the finite element method in handling singularities on the surface of grounded conductors. The fast multipole algorithm approximates the far-field action through multipole expansion and local expansion, reducing computational complexity from... Reduce to , The number of unknowns is represented by the following: The ground potential spatial distribution matrix is ​​a two-dimensional array containing the coordinates of each measuring point on the ground surface and its corresponding potential value. The number of rows represents the number of measuring points, and the number of columns is 4. The first 3 columns represent the three-dimensional coordinates of the measuring points, and the 4th column represents the potential value. The current density distribution matrix is ​​a two-dimensional array containing the location of each conductor segment of the grounding grid and its current density value. The unit of current density is A / m². .

[0036] The current distribution uniformity entropy value is an index that quantifies the degree of current distribution uniformity based on information entropy theory. The calculation method involves normalizing the current values ​​of each conductor segment in the grounding grid to a probability distribution, taking the logarithm of the probability, multiplying it by the probability, summing the results, and taking the negative value. A larger current distribution uniformity entropy value indicates a more uniform current distribution; the entropy value reaches its maximum when the distribution is completely uniform. The current concentration Gini coefficient is an index that measures the degree of current distribution non-uniformity, ranging from 0 to 1. A larger value indicates a more severe current concentration phenomenon. A Gini coefficient less than 0.3 indicates a relatively uniform distribution, while a value greater than 0.6 indicates severe concentration. The entropy threshold is determined based on the grounding grid size; for areas less than 2000... The grounding grid should be 2.5 to 3.0 for areas larger than 2000. The grounding grid is set to 3.5 to 4.5. The Gini coefficient threshold is uniformly set to 0.4. The current concentration area is the region where the conductor segment with a current density value greater than 3 times the average current density value is located in the current density distribution matrix. The local overheating risk index comprehensively considers the peak current density, duration, and conductor heat dissipation conditions, and the calculation formula is as follows: Divide the peak current density by the conductor's allowable current density to obtain the current overload coefficient; divide the impact duration by the conductor's thermal time constant to obtain the time coefficient; multiply the current overload coefficient by the time coefficient and then multiply by the heat dissipation correction factor to obtain the local overheating risk index. The heat dissipation correction factor is determined based on the conductor's burial depth and the thermal resistance of the surrounding soil, and its value ranges from 0.5 to 2.0. The peak current density is the maximum current density in the current concentration area.

[0037] The Analytic Hierarchy Process (AHP) decomposes decision problems into a hierarchical structure of objective, criterion, and alternative layers. It's a multi-criteria decision-making method that constructs a judgment matrix through pairwise comparisons and calculates weights. A weight allocation is considered reasonable when the consistency ratio of the judgment matrix is ​​less than 0.1. The initial values ​​of the comprehensive weight coefficients are set based on empirical values ​​of the influence of each indicator on mid-frequency performance. The comprehensive weight coefficient corresponding to the impedance peak is 0.25, the comprehensive weight coefficient corresponding to the resonant frequency distribution parameter is 0.20, the comprehensive weight coefficient corresponding to the ground potential gradient is 0.25, the comprehensive weight coefficient corresponding to the current distribution uniformity entropy value is 0.15, and the comprehensive weight coefficient corresponding to the local overheating risk index is 0.15. The impedance threshold is determined based on the rated voltage level of the grounding grid: 20Ω for 10kV systems, 10Ω for 35kV systems, and 5Ω for 110kV and above systems. The sensitive frequency band refers to the frequency range where the filter control system and secondary equipment are susceptible to interference, mainly including two bands: 800Hz to 3kHz and 8kHz to 15kHz. The ground potential gradient is the ratio of the potential difference to the distance between adjacent measuring points in the ground potential spatial distribution matrix. The gradient limit refers to the maximum allowable value of the ground potential gradient; to prevent step voltage hazards, the gradient limit is set between 50V / m and 100V / m. The safety scoring threshold is set at 75 points, and the comprehensive score of the intermediate frequency performance of the grounding grid is based on a percentage system. A score below the safety scoring threshold indicates a potential safety hazard in the intermediate frequency performance of the grounding grid.

[0038] The performance evaluation report includes five parts: basic information of the grounding grid, test results of various performance indicators, comprehensive score of the intermediate frequency performance of the grounding grid, analysis of weak points, and improvement suggestions. The analysis of weak points identifies the location and cause of performance deficiencies based on the scores of each indicator. The improvement suggestions propose measures such as increasing the grounding grid area, optimizing conductor layout, improving soil resistivity, and installing damping devices based on the analysis results. The weak point location information includes the spatial coordinates of the current concentration area and the locations of measurement points where the ground potential gradient exceeds the gradient limit.

[0039] As a further solution, the present invention also provides a computer-based method for forming a comprehensive evaluation system for the intermediate frequency performance of a filter field grounding system. The computer is equipped with a readable storage medium that stores program instructions. When the program instructions are run on the computer, they execute the aforementioned comprehensive evaluation method for the intermediate frequency performance of a filter field grounding system.

[0040] The specific implementation methods of the above steps are described in detail below.

[0041] The specific implementation of step S01 involves first arranging broadband impedance measurement devices at the outer boundary nodes, center nodes, and quarter-point positions of the filter field grounding grid. The placement positions are selected based on the symmetry of the grounding grid to reduce the number of measurement points. A multi-frequency synchronous injection technique is used to generate a composite sine wave containing 128 logarithmically spaced frequency points within the range of 0.5kHz to 10kHz as a frequency sweep excitation signal using a digital signal generator. This composite sine wave is amplified by a power amplifier and injected into the grounding grid. Simultaneously, current and voltage response signals are measured at the injection points. The response signals are then multiplied by the in-phase reference signal and the quadrature reference signal using a quadrature lock-in amplifier, and then passed through a low-frequency amplifier. The process involves filtering to obtain the in-phase and quadrature components at each frequency point. Preliminary impedance data is obtained by calculating the amplitude and phase of these components. Frequency-domain averaging of this preliminary impedance data involves arithmetic averaging of 16 consecutive measurements to reduce white noise. Time-domain adaptive filtering is then applied to the averaged impedance data using a minimum mean square error algorithm. This algorithm estimates the background noise statistical characteristics based on the first 10 sampling points and automatically adjusts the filter coefficients. The filter coefficient update step size is between 0.01 and 0.05. Finally, a wideband impedance spectrum curve with a signal-to-noise ratio greater than 40 dB is obtained. This wideband impedance spectrum curve includes the impedance amplitude and phase angle data at each frequency point. The purpose of these steps is to obtain basic impedance characteristic data of the grounding grid in the mid-frequency range. Compared to the point-by-point frequency sweeping method, multi-frequency synchronous injection technology significantly improves testing efficiency through parallel measurement. The quadrature lock-in amplifier utilizes correlation detection principles to extract weak signals from strong interference backgrounds. The dual processing of frequency-domain averaging and time-domain adaptive filtering ensures high signal-to-noise ratio and reliability of the measurement data.

[0042] The specific implementation of step S02 is as follows: First, a ground-penetrating radar is used to scan the ground grid by setting up survey lines with a spacing of 2m to 3m. The ground-penetrating radar emits electromagnetic pulses of 100MHz to 500MHz and records the arrival time and amplitude of the reflected waves. The depth of the interface between each soil layer is calculated based on the time difference of the reflected waves. The location of abrupt changes in the soil dielectric constant is identified based on the changes in the amplitude of the reflected waves, thereby obtaining the layered structure data. Then, resistivity imaging technology is used to arrange a Wenner quadrupole electrode array around the ground grid with a spacing of 2m between the electrodes. Current is injected into different electrode pairs in sequence, and the potential difference of the remaining electrode pairs is measured. The underground resistivity distribution is solved by a least-squares inversion algorithm. The inversion algorithm iterates 20 to 30 times, and the convergence criterion is that the relative residual is less than 5%. After obtaining the resistivity distribution data, a system based on fractional differential equations is established. A soil frequency-varying model is constructed, which uses Caputo-type fractional derivatives to describe the frequency dispersion characteristics of the soil medium. The broadband impedance spectrum curve obtained in step S01 is used as the objective function. A fractional-order particle swarm optimization algorithm is used to optimize the fractional-order parameters and relaxation time parameters of the soil frequency-varying model. The fractional-order particle swarm optimization algorithm initializes 100 particles, and the position vector of each particle contains fractional-order parameters and relaxation time parameters. The velocity update formula introduces a fractional order of 0.7 to 0.85 to enable the particles to have historical trajectory memory. After 50 to 100 iterations, the optimization is terminated when the global optimal fitness value changes by less than 0.1% for 10 consecutive generations. The optimized fractional-order parameters, relaxation time parameters, and the functional relationships of soil resistivity and dielectric constant with frequency are stored as a soil frequency domain equivalent parameter dataset. The purpose of these steps is to establish an accurate soil frequency-varying characteristic model to provide boundary conditions for subsequent electromagnetic field simulation. Ground-penetrating radar and resistivity imaging technologies complement each other to obtain soil structure and electrical parameters. Fractional differential equations can describe the non-Debye relaxation and memory effects of the soil medium. Compared with the standard particle swarm algorithm, the fractional particle swarm algorithm has a stronger global optimization capability, thereby improving the accuracy of parameter inversion.

[0043] The specific implementation of step S03 involves inputting the broadband impedance spectrum curve obtained in step S01 into a pre-trained frequency domain response identification model. The broadband impedance spectrum curve is first normalized to map the impedance amplitude and phase angle to the intervals of 0 to 1 and -1 to 1, respectively. The normalized data undergoes multi-scale convolution operations through the feature extraction layer of the frequency domain response identification model to extract spectral features. The four-layer one-dimensional convolutional neural network of the feature extraction layer sequentially extracts frequency domain features from coarse to fine granular. The convolution output is processed by a modified linear unit activation function with leakage and then input into a coarse spatiotemporal attention layer. The spatial attention module of the coarse spatiotemporal attention layer generates an attention weight matrix based on the current spatial receptive field parameter values. The attention weight matrix is ​​element-wise multiplied with the feature map to achieve selective enhancement for different frequency bands. The temporal attention module calculates the similarity matrix between frequency points through a self-attention mechanism and generates time-dimensional attention weights. The weighted features are then input into a three-layer fully connected neural network of the parameter prediction layer. The fully connected neural network outputs resonant frequency distribution parameters. The parameters are quality factor and modal damping. When calculating the resonant frequency offset function, the predicted resonant frequency output by the frequency domain response identification model is subtracted from the standard resonant frequency obtained by looking up the table based on the grounding grid area and down conductor length. This subtraction is then divided by the standard resonant frequency to obtain the relative deviation. The quality factor parameter is divided by the standard quality factor of 5 to 15 (midpoint 10) to obtain the normalized quality factor value. The impedance peak value in the broadband impedance spectrum curve obtained in step S01 is divided by the standard impedance peak value of 10Ω to 100Ω (midpoint 55Ω) to obtain the impedance. The normalized value, multiplied by 0.5, 0.3, and 0.2 respectively, and summed to obtain the resonant frequency offset function value. The spatial receptive field parameter is adjusted according to the interval to which the resonant frequency offset function value belongs: when the resonant frequency offset function value is less than 0.1, the spatial receptive field parameter is set to 3; when the resonant frequency offset function value is between 0.1 and 0.3, it is set to 7; when the resonant frequency offset function value is between 0.3 and 0.6, it is set to 13; and when the resonant frequency offset function value is greater than or equal to 0.6, it is set to 18. The purpose of these steps is to quickly and accurately predict the resonant characteristics of the grounding grid based on the measured broadband impedance spectrum curve. The frequency domain response identification model automatically extracts the complex mapping relationship between the impedance spectrum and the resonant parameters through deep learning, avoiding the simplification assumptions of traditional analytical methods. The coarse spatiotemporal resolution attention mechanism adaptively adjusts the model's focus range according to the resonant frequency offset, achieving a dynamic balance between accuracy and efficiency.

[0044] The specific implementation of step S04 involves establishing a three-dimensional electromagnetic field simulation model of the grounding grid based on the fractional-order parameters, relaxation time parameters, soil resistivity, and dielectric constant in the soil frequency domain equivalent parameter dataset generated in step S02. The simulation model models the grounding grid conductor according to its actual geometric dimensions. The soil region is set as a cuboid computational domain with dimensions 5 times the grounding grid size and a height of 50m. Radial basis function collocation is used to arrange collocation points within the computational domain. These collocation points are densely arranged near the grounding conductor with a spacing of 0.1m to 0.5m, and sparsely arranged in areas far from the conductor with a spacing of 2m to 5m. A quadratic radial basis function is selected as the interpolation basis function. The shape parameters of the quadratic function are determined by leaving... Cross-validation determined the value range to be 1 to 3. Radial basis functions were used to approximate the potential distribution and establish a system of linear equations. When using the boundary element method at the grounding conductor boundary, the conductor surface was discretized into triangular elements with side lengths of 0.05m to 0.2m. Green's function was used to establish boundary integral equations to accurately consider the potential singularity of the conductor surface. For the far-field region more than 5 times the mesh size away from the grounding conductor, a fast multipole algorithm was applied. This algorithm partitioned the computational domain into an octree structure with 5 to 8 layers. Multipole expansion and local expansion reduced the computational complexity of the far-field effect. The preconditioned conjugate gradient method was used to solve the system of linear equations, and the iterative convergence criterion was that the residual norm was less than 3. After obtaining the potential values ​​of each node, the ratio of the potential difference to the distance between adjacent nodes is calculated to obtain the spatial distribution matrix of the ground potential. The current density distribution matrix is ​​then obtained by calculating the current density distribution based on the potential gradient and soil conductivity. The purpose of these steps is to obtain the potential and current distribution of the grounding grid under medium-frequency impulse conditions through numerical simulation. The radial basis function collocation method, as a meshless method, avoids the mesh generation difficulties of the finite element method. The boundary element method's accurate handling of conductor surface singularities ensures the accuracy of the boundary conditions, and the fast multipole algorithm significantly reduces the computation time for large-scale problems.

[0045] The specific implementation of step S05 involves extracting the current density values ​​of each conductor segment from the current density distribution matrix obtained in step S04. When normalizing the current density values ​​of all conductor segments into a probability distribution, the sum of current densities is first calculated. Then, the current density values ​​of each conductor segment are divided by the sum of current densities to obtain probability values. When calculating the entropy value of current distribution uniformity, the natural logarithm of each probability value is taken, multiplied by the corresponding probability value, and the sum is taken over all conductor segments, with the result being negative. It is then determined whether the entropy value of current distribution uniformity is less than the entropy threshold determined based on the area of ​​the grounding grid. For areas less than 2000... The entropy threshold for the grounding grid is set to 2.7, for areas larger than 2000. The entropy threshold for the grounding grid is set to 4.0. When calculating the Gini coefficient for current concentration, the current density values ​​of each conductor segment are first arranged in ascending order. A Lorentz curve is constructed, and the ratio of the area under the curve to the area under the diagonal is calculated. Twice this ratio minus 1 gives the Gini coefficient for current concentration. It is then determined whether the Gini coefficient is greater than the Gini coefficient threshold of 0.4. If the entropy value for current distribution uniformity is less than the entropy threshold or the Gini coefficient for current concentration is greater than the Gini coefficient threshold, a conductor segment with a current density value greater than three times the average current density value is found in the current density distribution matrix. This conductor segment is marked as a current concentration area. The maximum current density value within the current concentration area is extracted as the peak current density. The peak current density is then divided by the conductor's allowable current density of 1000A / m³. Up to 5000A / The current overload coefficient is obtained. The duration of the medium-frequency impulse (0.01s to 0.1s) is divided by the conductor thermal time constant (10s to 100s) to obtain the time coefficient. A heat dissipation correction factor is determined by referring to a table based on the conductor burial depth (0.5m to 2m) and soil thermal resistance (1℃·m / W to 3℃·m / W). The current overload coefficient is multiplied by the time coefficient, and then multiplied by the heat dissipation correction factor to obtain the local overheating risk index. The purpose of these steps is to quantitatively assess the uniformity of the grounding grid current distribution and identify potential overheating risk areas. The current distribution uniformity entropy value, based on information entropy theory, can comprehensively reflect the spatial distribution pattern of the current. The current concentration Gini coefficient, using income inequality measurement methods from economics, intuitively characterizes the degree of current concentration. The local overheating risk index comprehensively considers the degree of current overload, the duration of action, and heat dissipation conditions to provide a quantitative basis for conductor thermal stability assessment.

[0046] The specific implementation of step S06 involves using the impedance peak value from the broadband impedance spectrum curve obtained in steps S01 to S05, the resonant frequency distribution parameters output in step S03, the ground potential spatial distribution matrix calculated in step S04, the current distribution uniformity entropy value calculated in step S05, and the local overheating risk index as evaluation indicators. A three-layer hierarchical structure is constructed using the analytic hierarchy process (AHP). The target layer is the comprehensive score of the grounding grid's intermediate frequency performance. The criteria layer includes five dimensions: impedance characteristics, resonant characteristics, potential distribution, current uniformity, and thermal stability. The scheme layer consists of the actual measured values ​​of each indicator. A pairwise comparison judgment matrix for each indicator in the criteria layer is established through expert scoring. The eigenvector corresponding to the largest eigenvalue of the judgment matrix is ​​calculated and normalized to obtain the initial value of the comprehensive weight coefficient. The initial values ​​of the comprehensive weight coefficient are 0.25, 0.20, 0.25, 0.15, and 0.15, respectively. The system then determines whether the impedance peak value of the broadband impedance spectrum curve is greater than the impedance threshold determined based on the rated voltage level of the grounding grid, and whether the resonant frequency in the resonant frequency distribution parameters falls within 800 Hz. In the sensitive frequency bands from z to 3kHz or 8kHz to 15kHz, the potential gradient is calculated by the ratio of the potential difference to the distance between adjacent measuring points in the spatial distribution matrix of the ground potential to obtain the ground potential gradient and whether it exceeds the gradient limit of 75V / m. When any of the above conditions are met, the comprehensive weight coefficient of the corresponding index is increased by 40% of the original value. After normalizing each index, it is multiplied by the adjusted comprehensive weight coefficient and summed to obtain the comprehensive score value of the intermediate frequency performance of the grounding grid. It is determined whether the comprehensive score value of the intermediate frequency performance of the grounding grid is lower than the safety score threshold of 75 points. If it is lower than the safety score threshold, a performance evaluation report is generated. The performance evaluation report includes basic information on the geometric and material parameters of the grounding grid, the measured values ​​and scores of each index, the comprehensive score value of the intermediate frequency performance of the grounding grid, the weak link analysis based on the three lowest-scoring indicators to identify the weak link and analyze the physical causes, and specific technical improvement suggestions are proposed for the weak link analysis. The improvement suggestions include increasing the grounding grid area, adjusting the conductor spacing, replacing low-resistivity soil, and installing parallel resistor damping, etc. The purpose of these steps is to comprehensively evaluate the intermediate frequency performance of the grounding grid using multi-dimensional indicators and output a diagnostic report. The analytic hierarchy process (AHP) decomposes the complex multi-indicator decision-making problem into a hierarchical structure, making the evaluation process systematic. Dynamically adjusting the comprehensive weight coefficients can highlight the importance of indicators that exceed the standards. The performance evaluation report provides decision support for the optimization and transformation of the grounding system.

[0047] It should be noted that the key technical ideas of this invention include the following three aspects. The first key technical idea is to use multi-frequency synchronous injection technology combined with quadrature lock-in amplifier to achieve high-precision wideband impedance measurement. Traditional methods use point-by-point frequency sweeping, which is inefficient and easily affected by field interference, resulting in insufficient signal-to-noise ratio in the mid-frequency band. This invention injects a composite excitation signal containing multiple frequency components at the same time and uses the correlation detection principle of quadrature lock-in amplifier to extract the response of each frequency point from the strong noise background. Compared with traditional methods, the measurement time is shortened to one-tenth of the original and the signal-to-noise ratio is improved to more than 40dB, providing high-quality basic data for subsequent analysis. The second key technical approach is to establish a soil frequency-varying model based on fractional differential equations and use a fractional particle swarm optimization algorithm for parameter inversion. Traditional soil models using constant resistivity or integer-order Debye models cannot accurately describe the frequency dispersion and memory effect of the soil medium. This invention introduces fractional derivatives to characterize the fractal properties of soil microstructure and utilizes the global optimization capability of the fractional particle swarm optimization algorithm to invert model parameters from broadband impedance measurement data. Compared with integer-order models, the prediction accuracy of soil electrical properties in the mid-frequency band is improved by 40%, significantly improving the accuracy of boundary conditions in electromagnetic field simulation. The third key technical approach is to design a frequency domain response identification model and introduce a coarse spatiotemporal resolution attention mechanism to achieve rapid prediction of resonance characteristics. Traditional methods rely on lumped parameter simplification analysis or full-wave simulation, the former lacking in accuracy and the latter involving large computational loads. This invention establishes an end-to-end mapping relationship between impedance spectrum and resonance parameters through deep learning. The coarse spatiotemporal resolution attention mechanism adaptively adjusts the model's focus range according to the resonance frequency offset. When the offset is large, the receptive field is expanded to capture wide-band coupling characteristics, and when the offset is small, local details are focused to improve prediction accuracy. Under the premise of ensuring that the average relative error is less than 5%, the computation time is shortened to one-thousandth of that of traditional full-wave simulation. The synergistic effect of the three key technical approaches mentioned above is that high-precision broadband impedance measurement provides reliable input data for soil frequency-varying model parameter inversion and resonance characteristic prediction; accurate soil frequency-varying model establishes high-fidelity boundary conditions for electromagnetic field simulation; and rapid resonance characteristic prediction avoids time-consuming full-wave simulation. The three support each other to form a complete mid-frequency performance evaluation chain. Compared with existing technologies that only focus on power frequency or lightning impulse conditions, this invention establishes a systematic evaluation system for the unique frequency response characteristics and resonance risks of mid-frequency oscillations, solving the technical problem of lacking an effective method for evaluating the mid-frequency performance of filter field grounding systems.

[0048] It should be noted that this invention also solves the following technical problem: traditional grounding system assessment methods struggle to accurately identify current concentration areas and local overheating risk points. While existing technologies can calculate the current distribution of the grounding grid through finite element simulation, the lack of a quantitative evaluation system for current distribution uniformity and the failure to consider the impact of conductor heat dissipation conditions and impact duration on overheating risk prevent effective identification of weak points in the grounding system. This invention introduces a current distribution uniformity entropy value based on information entropy theory and a Gini coefficient measuring the degree of distribution non-uniformity to quantitatively assess current concentration phenomena. Current concentration areas are marked when the entropy value is less than a threshold or the Gini coefficient is greater than a threshold. Furthermore, this invention calculates a local overheating risk index for current concentration areas. This index comprehensively considers the ratio of peak current density to allowable value, the ratio of impact duration to conductor thermal time constant, and a heat dissipation correction factor determined based on conductor burial depth and soil thermal resistance, thereby accurately assessing local overheating risk and providing a quantitative basis for the optimized design of grounding systems.

[0049] Specifically, the principle of this invention is as follows: This invention can solve the technical problem of inaccurate mid-frequency performance evaluation in filter field grounding systems. Its principle lies in establishing a complete technical chain from broadband measurement to frequency-varying modeling and then to comprehensive evaluation. First, the multi-frequency synchronous injection technology acquires impedance data at multiple frequencies in parallel during a single measurement. The quadrature lock-in amplifier effectively suppresses noise interference through correlation operations, ensuring the accuracy of the broadband impedance spectrum and providing reliable frequency domain characteristic data for subsequent analysis. Second, the soil model based on fractional differential equations describes the memory effect of charge diffusion through Caputo-type derivatives. The fractional particle swarm optimization algorithm inverts the fractional order and relaxation time parameters from the measured impedance spectrum, enabling the established three-dimensional electromagnetic field simulation model to realistically reflect the frequency-dependent characteristics of the soil, thereby accurately calculating the ground potential and current density distribution under mid-frequency impacts. Third, the frequency domain response identification model adaptively adjusts the degree of attention to different frequency bands through a coarse spatiotemporal attention mechanism, dynamically adjusting the spatial receptive field parameters according to the resonant frequency offset. This captures both the coupling characteristics of the broadband band and focuses on local details, achieving accurate extraction of resonant parameters. Finally, the degree of current dispersion is quantified by the entropy value of current distribution uniformity, the Gini coefficient is used to identify concentration phenomena, and the local overheating risk index integrates the peak current density and heat dissipation conditions. The analytic hierarchy process (AHP) assigns weights to each indicator based on their impact on safety and dynamically adjusts them, forming a scientific and reasonable comprehensive scoring mechanism. This multi-dimensional and multi-scale evaluation system can comprehensively and accurately reflect the true performance of the grounding system under medium-frequency impulse conditions, conforming to the physical laws of electromagnetic transient processes.

[0050] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0051] The specific implementation of step S01 is as follows: Wideband impedance measurement devices are arranged at multiple nodes of the filter field grounding grid. A multi-frequency synchronous injection technique is used to inject a swept-frequency excitation signal in the range of 0.5kHz to 10kHz into the grounding grid. The response signal is extracted by an orthogonal lock-in amplifier and then processed by frequency domain averaging and time domain adaptive filtering to obtain the wideband impedance spectrum curve. The multi-frequency synchronous injection technique injects multiple sinusoidal excitation signals of different frequencies into the grounding grid simultaneously, resulting in a superimposed waveform. The statement is as follows: ; In the formula, The number of injection frequency points is set to 200 to 500. For the first The amplitude of each frequency component, in amperes (A); For the first The frequency of each frequency component is expressed in Hz. Time, in seconds; For the first The initial phase of each frequency component, in rad; To superimpose the excitation signal, the unit is Am. The response signal is decomposed into frequency components using Fourier transform, enabling parallel impedance measurements at multiple frequencies. An orthogonal lock-in amplifier extracts frequency components by performing correlation operations between the reference signal and the signal under test, and amplitude and phase information are separated using orthogonal demodulation techniques. Frequency domain averaging processes the spectral data obtained from multiple measurements using an arithmetic mean, resulting in the impedance spectrum after frequency domain averaging. The statement is as follows: ; In the formula, For frequency The average impedance at point, in units of ; The average number of times is taken as 8 to 16. For the first The measurement at frequency The impedance value at the location, in units of The time-domain adaptive filtering process uses the minimum mean square error criterion to update the filter coefficients, which has a strong ability to suppress non-stationary interference.

[0052] The specific implementation of step S02 is as follows: Ground-penetrating radar and resistivity imaging technology are used to acquire layered structure data and resistivity distribution data of the soil surrounding the grounding grid. A soil frequency-varying model based on fractional differential equations is established. A fractional-order particle swarm optimization algorithm is used to invert the fractional-order parameters and relaxation time parameters of the soil frequency-varying model from the broadband impedance spectrum curve obtained in step S01, generating a soil frequency-domain equivalent parameter dataset. The ground-penetrating radar operates in the frequency range of 10MHz to 1000MHz, with a detection depth of 5m to 30m and a resolution of 0.1m to 0.5m. The resistivity imaging technology measures the electrode spacing from 1m to 5m, with an imaging depth of 0.3 to 0.5 times the electrode spacing. The soil frequency-varying model based on fractional differential equations uses Caputo-type fractional derivatives to describe the memory effect of charge diffusion and polarization relaxation in the soil medium, and the soil complex resistivity... The statement is as follows: ; In the formula, Angular frequency Soil resistivity, in units of ; DC resistivity, in units of ; The polarization coefficient is dimensionless. The imaginary unit satisfies ; Angular frequency, unit: ; The relaxation time parameter is expressed in seconds and has a range of values. s to s; This is a fractional order parameter, dimensionless, with values ​​ranging from 0 to 2. The fractal dimension reflects the microscopic pore structure of soil; the closer it is to 1, the more homogeneous the soil structure. The fractional-order particle swarm optimization algorithm introduces a fractional-order differential operator into the velocity update formula of the standard particle swarm optimization algorithm. The particle velocity update formula is expressed as follows: ; In the formula, For the first The particle in the first The speed of each iteration, in units of ; Maximum speed, unit: ; The inertial weight is dimensionless and ranges from 0.4 to 0.9. For the first The particle in the first The speed of each iteration, in units of ; and The learning factor is dimensionless and typically takes a value of 2. and A random number between 0 and 1, dimensionless; For the first The historical best position of each particle, in meters; For the first The particle in the first The position of the next iteration, in meters; This represents the maximum position value, in meters (m). The globally optimal position, in meters (m). These are fractional weighting coefficients, dimensionless, with values ​​ranging from 0.1 to 0.5. for Fractional differential operators are used to apply velocity. First-order differential operations; It is a fractional derivative order, dimensionless, and its convergence performance is optimal when the order is between 0.6 and 0.9. Number the particles; The iteration number is given. The soil frequency domain equivalent parameter dataset includes fractional-order parameters, relaxation time parameters, soil resistivity, and soil dielectric constant values ​​at different frequencies.

[0053] The specific implementation of step S03 is as follows: The broadband impedance spectrum curve obtained in step S01 is input into the frequency domain response identification model for processing. The frequency domain response identification model outputs resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. The resonant frequency offset function value is calculated, and the spatial receptive field parameters of the coarse spatiotemporal resolution attention mechanism of the frequency domain response identification model are adjusted according to the interval to which the resonant frequency offset function value belongs. The frequency domain response identification model includes a feature extraction layer, a coarse spatiotemporal attention layer, and a parameter prediction layer. The feature extraction layer consists of four one-dimensional convolutional neural networks with kernel sizes of 16, 32, 64, and 128, a stride of 2, and a modified linear unit with leakage as the activation function. Multi-scale feature extraction is performed on the input broadband impedance spectrum curve. The coarse spatiotemporal attention layer includes a spatial attention module and a temporal attention module. The spatial attention module controls the degree of attention to different frequency bands through the spatial receptive field parameters. The temporal attention module uses a self-attention mechanism to capture the long-range dependence of the broadband impedance spectrum curve between different frequency points. The number of attention heads is 8, and the dimension of each attention head is 64. The parameter prediction layer consists of three fully connected neural network layers with 256, 128, and 64 neurons respectively, outputting resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. The resonant frequency offset function... The statement is as follows: ; In the formula, The value of the resonant frequency offset function is dimensionless; The resonant frequency is predicted in Hz. The standard resonant frequency is expressed in Hz. For predicting quality factors, dimensionless; The standard quality factor is dimensionless and ranges from 5 to 15. Peak impedance, in units of ; Standard impedance peak value, in units of The value range is 10. Up to 100 Standard resonant frequency Based on the grounding grid area and lead wire length Determined by looking up the table, among which The unit is , The unit is meters (m). When the resonant frequency offset function value... Spatial receptive field parameters when belonging to different intervals The adjustment rules are as follows: when hour Take 2 to 4, when hour Take 5 to 10, when hour Take 11 to 15, when hour Take 16 to 20, of which This is a dimensionless parameter. Quality factor The sharpness of the resonance peak is characterized by the following formula: ; In the formula, The resonant frequency is expressed in Hz. The half-power bandwidth of the resonant peak, in Hz. Modal damping parameters. It reflects the rate of energy dissipation, is dimensionless, and ranges from 0.01 to 0.1.

[0054] The specific implementation of step S04 is as follows: Based on the soil frequency domain equivalent parameter dataset generated in step S02, a three-dimensional electromagnetic field simulation model of the grounding grid is established. The radial basis function collocation method is used to discretize and solve the surface potential distribution. The boundary element method is used to handle potential singularities at the grounding conductor boundary. The fast multipole algorithm is applied to accelerate the calculation in the far-field region to obtain the spatial distribution matrix of the ground potential and the current density distribution matrix under the medium-frequency impulse condition. The radial basis function collocation method uses radial basis functions as interpolation basis functions to obtain the surface potential... The approximate solution is expressed as follows: ; In the formula, For spatial points The electrical potential at a point, expressed in volts (V). This is the reference potential, in volts (V), with an empirical value of 100V. The number of points; The coefficients to be determined are dimensionless. These are radial basis functions, dimensionless, and employ multiple quadratic form functions. or Gaussian function ; For the first The coordinates of the collocation points are in meters. , , These are the three-dimensional coordinates of a point in space, in meters (m). This is the Euclidean distance, in meters (m). The characteristic length, in meters, is determined based on the grounding grid size. The distance is normalized and dimensionless. The boundary element method (BEM) transforms the internal field problem into a boundary integral equation solution using the Green's function, and its accuracy in handling the singularities of the grounded conductor surface is superior to the finite element method. The fast multipole algorithm approximates the far-field action through multipole expansion and local expansion, reducing computational complexity from... Reduce to ,in The number of unknowns. Spatial distribution matrix of ground potential. This is a two-dimensional array containing the coordinates of each measuring point on the Earth's surface and its corresponding potential value, with the number of rows equal to the number of measuring points. The matrix has four columns; the first three columns represent the three-dimensional coordinates of the measuring point, and the fourth column represents the potential value. (Current density distribution matrix) This is a two-dimensional array containing the location and current density values ​​of each conductor segment of the grounding grid, with the current density unit being... .

[0055] The specific implementation of step S05 is as follows: Calculate the current distribution uniformity entropy value and the current concentration Gini coefficient from the current density distribution matrix obtained in step S04. When the current distribution uniformity entropy value is less than the entropy threshold or the current concentration Gini coefficient is greater than the Gini coefficient threshold, mark the current concentration area and calculate the local overheating risk index of the current concentration area. Current distribution uniformity entropy value The statement is as follows: ; In the formula, The entropy value of the uniformity of current distribution is dimensionless; This refers to the number of conductor segments in the grounding grid. For the first The normalized probability of the current in a conductor segment is dimensionless and is calculated as follows: ; For the first The current value of a conductor segment, expressed in amperes (A). Conductor segment numbering. Current concentration Gini coefficient. The value of is between 0 and 1. The larger the value, the more severe the current concentration phenomenon. The calculation formula is as follows: ; In the formula, For the first The current value of a conductor segment, expressed in amperes (A). Number the conductor segments; This is the average current value, in amperes (A), calculated as follows: Entropy threshold Determined based on the grounding grid size, for areas smaller than 2000 The grounding grid should be 2.5 to 3.0 for areas larger than 2000. The grounding grid is dimensionless, ranging from 3.5 to 4.5. Gini coefficient threshold. The value is uniformly set to 0.4, dimensionless. The current concentration region is the area in the current density distribution matrix where the current density value is more than three times the average current density value. The unit is Local overheating risk index The statement is as follows: ; In the formula, This is a dimensionless index representing the risk of localized overheating. Peak current density, in units of ; The allowable current density of the conductor, in units of ; The duration of the impact is expressed in seconds (s). The thermal time constant of the conductor is expressed in seconds (s). This is a dimensionless heat dissipation correction factor, determined by the conductor burial depth. Thermal resistance of surrounding soil It is determined that the value ranges from 0.5 to 2.0, where... The unit is m. The unit is .

[0056] The specific implementation of step S06 is as follows: Based on the broadband impedance spectrum curve, resonant frequency distribution parameters, ground potential spatial distribution matrix, current distribution uniformity entropy value, and local overheating risk index obtained in steps S01 to S05, the comprehensive weighting coefficient of each indicator is calculated using the analytic hierarchy process (AHP). When the impedance peak value of the broadband impedance spectrum curve is greater than the impedance threshold, or the resonant frequency in the resonant frequency distribution parameters falls into a sensitive frequency band, or the ground potential gradient in the ground potential spatial distribution matrix exceeds the gradient limit, the comprehensive weighting coefficient of the corresponding indicator is increased by 30% to 50%. The weighted summation of each indicator yields the comprehensive score value of the grounding grid's intermediate frequency performance. When the comprehensive score value of the grounding grid's intermediate frequency performance is lower than the safety score threshold, a performance evaluation report containing weak point location information and improvement suggestions is generated. (Grounding grid intermediate frequency performance comprehensive score value) The statement is as follows: ; In the formula, The comprehensive score for the intermediate frequency performance of the grounding grid is given on a 100-point scale. , , , , The comprehensive weighting coefficients are dimensionless, with initial values ​​of 0.25, 0.20, 0.25, 0.15, and 0.15, respectively, satisfying the following conditions: ; The peak impedance score is given on a 100-point scale. The resonant frequency index is scored out of 100. The score for the ground potential gradient index is based on a percentage system. The score is based on the uniformity of current distribution, on a percentage basis. The score is based on a percentage scale for the local overheating risk index. (Ground potential gradient) The potential difference between adjacent measuring points in the spatial distribution matrix of ground potential is expressed by the following formula: ; In the formula, For the first The potential value at each measuring point, in V; For adjacent The potential value at each measuring point, in V; For the first and the The distance between measuring points is in meters (m). The unit is ; Number the measurement points; Adjacent measuring points are numbered. The Analytic Hierarchy Process (AHP) decomposes the decision problem into a hierarchical structure of objective, criterion, and alternative layers. A judgment matrix is ​​constructed through pairwise comparisons, and weights are calculated. A weight allocation is considered reasonable when the consistency ratio of the judgment matrix is ​​less than 0.1. Impedance threshold. Determined based on the rated voltage level of the grounding grid; for a 10kV system, 20 is used. For the 35kV system, take 10. For 110kV and above systems, take 5. The sensitive frequency bands refer to the frequency ranges where the filter control system and secondary equipment are susceptible to interference, mainly including two bands: 800Hz to 3kHz and 8kHz to 15kHz. Gradient limits. The maximum allowable ground potential gradient is set to 50. Up to 100 The safety score threshold is set at 75 points; a score below this threshold indicates a potential safety hazard in the intermediate frequency performance of the grounding grid. The performance evaluation report includes five parts: basic information about the grounding grid, test results for various performance indicators, a comprehensive score for the intermediate frequency performance of the grounding grid, analysis of weaknesses, and improvement suggestions. The analysis of weaknesses identifies the location and causes of performance deficiencies based on the scores of each indicator. The improvement suggestions address the analysis by proposing measures such as increasing the grounding grid area, optimizing conductor arrangement, improving soil resistivity, and installing damping devices.

[0057] To better understand and implement this invention, a specific application scenario of the invention is provided below as Example 2: To verify the effectiveness of the invention, technicians built a test environment to conduct a comprehensive evaluation of the intermediate frequency performance of a filter field grounding system in a 220kV substation. The filter field grounding grid has a rectangular structure with an area of ​​2400 square meters. The grounding grid is 60m long and 40m wide. The main grid is made of 60mm×6mm galvanized flat steel with a mesh size of 10m×10m. A ring of equipotential bonding tape is laid around the perimeter, and there are four down conductors connected to the filter equipment. The grounding grid is buried at a depth of 0.8m, and the surrounding soil has a three-layer structure: a 2m thick layer of sandy clay on the surface, a 5m thick layer of silty clay in the middle, and a bottom layer of strongly weathered rock.

[0058] Technicians first deployed broadband impedance measurement devices at nine locations on the grounding grid: the four corner points, the center point, and the four midpoints of the sides. A multi-frequency synchronous injection technique was used to inject a composite sine wave with 128 logarithmically spaced frequency points, ranging from 0.5 kHz to 10 kHz, into the grounding grid. This composite sine wave was generated by a digital signal generator and amplified by a power amplifier to a peak current of 50 A. The injection point was chosen at the center of the grounding grid. During the measurement, a quadrature lock-in amplifier synchronously acquired current and voltage signals. Each measurement point was measured 16 times consecutively. After frequency domain averaging and time domain adaptive filtering, a broadband impedance spectrum curve was obtained. The broadband impedance spectrum curve showed three distinct resonant peaks in the range of 200 Hz to 5 kHz, with peak impedances of 45 Ω, 62 Ω, and 38 Ω, corresponding to frequencies of 850 Hz, 2.3 kHz, and 9.2 kHz, respectively.

[0059] Technicians then used a ground-penetrating radar (GPR) to conduct electromagnetic wave scans along the perimeter of the grounding grid, with survey lines spaced 2.5m apart. The GPR operated at a frequency of 200MHz and had a detection depth of 15m. Based on the time difference of reflected waves, the thickness of the topsoil was calculated to be 1.8m to 2.2m, and the thickness of the middle soil layer was 4.5m to 5.5m, which largely matched the actual geological data. Technicians then used resistivity imaging technology to deploy a Wenner quadrupole array around the grounding grid, with electrodes spaced 2m apart, consisting of 80 electrodes arranged in a linear array. The measured resistivity distribution data showed that the resistivity of the topsoil was 120Ω·m to 180Ω·m, the middle soil resistivity was 60Ω·m to 90Ω·m, and the bottom rock resistivity was 300Ω·m to 500Ω·m. This resistivity distribution data is shown in Table 1.

[0060] Table 1. Soil stratified resistivity measurement data

[0061] Technicians established a soil frequency variation model based on fractional differential equations and used a fractional particle swarm optimization (PSO) algorithm to invert model parameters from broadband impedance spectrum curves. The PSO algorithm initialized with 100 particles and converged after 80 iterations, yielding a fractional order parameter of 0.85 for the surface soil and a relaxation time parameter of [missing value]. s, the fractional order parameter of the middle soil layer is 0.92, and the relaxation time parameter is s, the fractional order parameter of the underlying rock is 0.65, and the relaxation time parameter is The soil frequency domain equivalent parameter dataset generated based on the parameters contains soil resistivity and dielectric constant values ​​at various frequency points in the range of 0.5 kHz to 10 kHz.

[0062] Technicians input the broadband impedance spectrum curve into a pre-trained frequency domain response identification model, which outputs resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. For example... Figure 2 As shown, the hardware system of the frequency domain response identification model includes a data acquisition module, a signal processing module, a model calculation module, and a result output module. The model outputs the following: first-order resonant frequency of 865Hz, quality factor of 18.5, and modal damping parameter of 0.027; second-order resonant frequency of 2.28kHz, quality factor of 22.3, and modal damping parameter of 0.022; and third-order resonant frequency of 9.35kHz, quality factor of 15.6, and modal damping parameter of 0.032. (The last sentence appears to be incomplete and possibly refers to a technical description of a grounding grid area of ​​2400.) With a total lead-out length of 48m, the standard resonant frequency is found to be 900Hz from a table. The calculated relative deviation between the predicted resonant frequency of 865Hz and the standard resonant frequency is 0.039. The quality factor parameter 18.5 divided by the standard quality factor 10 yields a normalized quality factor value of 1.85. The peak impedance of 45Ω divided by the standard peak impedance of 55Ω yields a normalized impedance value of 0.818. Multiplying these three values ​​by weighting coefficients of 0.5, 0.3, and 0.2 respectively, and then summing them, yields a resonant frequency offset function value of 0.338. Since the resonant frequency offset function value falls within the range of 0.3 to 0.6, the technicians adjusted the spatial receptive field parameter of the frequency domain response identification model to 13.

[0063] Technicians established a three-dimensional electromagnetic field simulation model of the grounding grid based on a soil frequency domain equivalent parameter dataset. The computational domain of the simulation model is a cuboid with a length of 300m, a width of 200m, and a height of 50m. Radial basis function collocation was used to arrange collocation points within the computational domain, with a spacing of 0.2m near the grounding conductor and 3m away from the conductor, for a total of 15,000 collocation points. The shape parameter of the multiple quadratic radial basis function was determined to be 2.1 through leave-one-out cross-validation. The boundary element method was used at the grounding conductor boundary, discretizing the conductor surface into triangular elements with a side length of 0.1m, for a total of 2,400 elements. The fast multipole algorithm was applied to the far-field region beyond 300m from the center of the grounding grid, with an octree layer count of 6. The preconditioned conjugate gradient method was used to solve the linear equations, and after 850 iterations, the residual norm decreased to [value missing]. The convergence condition is met. The simulated spatial distribution matrix of ground potential contains the coordinates and potential values ​​of 400 measuring points on the Earth's surface, such as... Figure 3As shown, the spatial distribution of ground potential exhibits a decreasing trend outward from the grounding grid. At a distance of 10m from the boundary of the grounding grid, the ground potential drops to 50% of the peak value, and at a distance of 30m, it drops to 10% of the peak value. The ground potential gradient is obtained by calculating the ratio of the potential difference between adjacent measuring points to the distance. The maximum ground potential gradient occurs at the northeast corner of the grounding grid, with a value of 92V / m, exceeding the gradient limit of 75V / m.

[0064] Technicians extract the current density values ​​for each conductor segment from the current density distribution matrix, such as... Figure 4 As shown, the current density distribution exhibits significant non-uniformity, with the current density in the conductor segments at the northeast and southwest corners of the grounding grid being significantly higher than in other locations. After normalizing the current density values ​​of all conductor segments to a probability distribution, the calculated entropy value for current distribution uniformity is 2.68, which is lower than that for areas larger than 2000. The entropy threshold of the grounding grid is 4.0. The calculated Gini coefficient for current concentration is 0.52, which is greater than the Gini coefficient threshold of 0.4. This result indicates a significant non-uniformity in the current distribution of the grounding grid. Technicians located conductor segments with current density values ​​greater than three times the average current density in the current density distribution matrix, marking two current concentration areas in the northeast and southwest corners. The peak current density in the northeast corner current concentration area is 3850 A / s. Divide by the conductor's allowable current density of 2500A / The current overload factor was 1.54. The time coefficient was 0.0011 when the medium frequency impact duration of 0.05s was divided by the conductor thermal time constant of 45s. Based on the conductor burial depth of 0.8m and the soil thermal resistance of 1.8℃·m / W, the heat dissipation correction factor was determined to be 1.35 by referring to the table. The local overheating risk index was calculated to be 0.0023.

[0065] Technicians used the analytic hierarchy process (AHP) to calculate the comprehensive weight coefficients for each indicator. After constructing a judgment matrix, the initial values ​​of the comprehensive weight coefficients were calculated as follows: peak impedance 0.25, resonant frequency distribution parameter 0.20, ground potential gradient 0.25, current distribution uniformity entropy 0.15, and local overheating risk index 0.15. Since the peak impedance of 45Ω is greater than the impedance threshold of 5Ω for a 220kV system, the first-order resonant frequency of 865Hz falls within the sensitive frequency range of 800Hz to 3kHz, and the ground potential gradient of 92V / m exceeds the gradient limit of 75V / m, the technicians increased the comprehensive weight coefficients corresponding to the peak impedance, resonant frequency distribution parameter, and ground potential gradient by 40% respectively. The adjusted comprehensive weight coefficients are shown in Table 2.

[0066] Table 2 Adjusted Comprehensive Weighting Coefficients

[0067] After normalizing each indicator, the technicians multiplied it by the adjusted comprehensive weighting coefficient and then summed the results. Figure 5 As shown, the comprehensive score for the intermediate frequency performance of the grounding grid was 68.5 points, lower than the safety score threshold of 75 points. Technicians generated a performance evaluation report, which pointed out that the weak points of the grounding grid mainly included three aspects: excessively high impedance peaks, resonant frequencies falling into sensitive frequency bands, and excessive ground potential gradients. The analysis of these weak points indicated that the excessively high impedance peaks were due to the high soil resistivity and the relatively insufficient grounding grid area; the resonant frequencies falling into sensitive frequency bands were due to improper matching of the distributed inductance and capacitance parameters formed by the grid structure and the down conductors; and the excessive ground potential gradients were due to uneven current distribution leading to excessively large local potential gradients. The improvement suggestions included adding a 20m wide auxiliary grounding strip around the grounding grid to expand the area to 3600m. In the areas where current is concentrated in the northeast and southwest corners, additional parallel ground electrodes are installed to disperse the current. Damping resistors are added at the main down conductors to shift the resonant frequency to outside the sensitive frequency band. The surface soil is treated to reduce the resistivity to below 100 Ω·m.

[0068] The technological advancements brought about by this invention compared to traditional methods are mainly reflected in three aspects. Traditional grounding system evaluation methods primarily target power frequency and lightning impulse conditions, using power frequency grounding resistance and impulse grounding resistance as evaluation indicators. These methods fail to reflect the frequency response characteristics and resonance risks of grounding systems in the mid-frequency range. In contrast, this invention, through wideband impedance spectrum measurement and resonant frequency distribution parameter identification, can comprehensively reveal the dynamic characteristics of grounding systems under mid-frequency oscillation impulses, providing a basis for the anti-interference design of filter field secondary systems. Traditional soil modeling methods employ constant resistivity models or simple Debye models, neglecting the frequency dispersion and memory effects of the soil medium, resulting in significant simulation errors in the mid-frequency range. However, the soil frequency-varying model established in this invention, based on fractional differential equations, can accurately describe the fractal characteristics and polarization relaxation processes of the soil microstructure, significantly improving the accuracy of electromagnetic field simulation. Traditional resonance characteristic analysis relies on simplified lumped parameter models or time-consuming full-wave simulations. The former lacks accuracy, while the latter is inefficient. The frequency domain response identification model designed in this invention establishes an end-to-end mapping between impedance spectrum and resonance parameters through deep learning. It introduces a coarse spatiotemporal resolution attention mechanism to achieve an adaptive balance between accuracy and efficiency. While ensuring prediction accuracy, it significantly shortens the calculation time, enabling mid-frequency performance evaluation to be quickly applied to engineering practice.

[0069] It should be noted that the variables involved in this invention are explained in detail in Tables 3 and 4.

[0070] Table 3. Variable Explanation Table (Part 1)

[0071] Table 4. Variable Explanation Table (Part Two)

[0072] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A comprehensive evaluation method for the mid-frequency performance of a filter field-grounding system, characterized in that, Wideband impedance measurement devices are deployed at grounding grid nodes to obtain wideband impedance spectrum curves. Soil stratification structure data and resistivity distribution data are acquired using ground-penetrating radar and resistivity imaging technology to establish a soil frequency variation model and generate a soil frequency domain equivalent parameter dataset. The wideband impedance spectrum curves are input into the frequency domain response identification model to output resonant frequency distribution parameters, quality factor parameters, and modal damping parameters. The spatial receptive field parameters of the coarse spatiotemporal resolution attention mechanism are adjusted according to the resonant frequency offset function value. A three-dimensional electromagnetic field simulation model of the grounding grid is established to obtain the spatial distribution matrix of ground potential and the current density distribution matrix. The entropy value of current distribution uniformity, the Gini coefficient of current concentration, and the local overheating risk index are calculated. The comprehensive weight coefficient is calculated using the analytic hierarchy process, and the weighted summation of each index is used to obtain the comprehensive score value of the intermediate frequency performance of the grounding grid. When the comprehensive score value is lower than the safety score threshold, the weak link location information is generated and output.

2. The method according to claim 1, characterized in that, The steps to obtain the wideband impedance spectrum curve are as follows: a multi-frequency synchronous injection technique is used to inject a sweep excitation signal in the range of 0.5kHz to 10kHz into the ground grid. The response signal is then extracted by an orthogonal lock-in amplifier and processed by frequency domain averaging and time domain adaptive filtering to obtain the wideband impedance spectrum curve.

3. The method according to claim 2, characterized in that, Multi-frequency synchronous injection technology refers to injecting superimposed waveforms of multiple sinusoidal excitation signals of different frequencies into the grounding grid at the same time, and then using Fourier transform to decompose the response signal into frequency components to achieve parallel measurement of impedance at multiple frequency points.

4. The method according to claim 3, characterized in that, A quadrature lock-in amplifier is a signal processing device that uses a reference signal and the signal under test to perform correlation operations to extract frequency components and separates amplitude and phase information through quadrature demodulation technology.

5. The method according to claim 4, characterized in that, Frequency domain averaging is a method that uses the arithmetic mean of spectrum data obtained from multiple measurements to reduce the impact of random noise. Time domain adaptive filtering uses the minimum mean square error criterion to update the filter coefficients.

6. The method according to claim 5, characterized in that, The soil frequency variation model uses fractional differential equations based on Caputo-type fractional derivatives to describe the memory effect of charge diffusion and polarization relaxation in the soil medium, and employs fractional particle swarm optimization to invert fractional order parameters and relaxation time parameters.

7. The method according to claim 6, characterized in that, The frequency domain response identification model includes a feature extraction layer, a coarse spatiotemporal attention layer, and a parameter prediction layer. The feature extraction layer is composed of a one-dimensional convolutional neural network to extract features at multiple scales from the broadband impedance spectrum curve.

8. The method according to claim 7, characterized in that, The coarse spatiotemporal attention layer includes a spatial attention module and a temporal attention module. The spatial attention module controls the degree of attention to different frequency bands through spatial receptive field parameters, while the temporal attention module uses a self-attention mechanism to capture long-range dependencies.

9. The method according to claim 8, characterized in that, The resonant frequency offset function value is calculated by multiplying the relative deviation between the predicted resonant frequency and the standard resonant frequency, the normalized value of the quality factor, and the normalized value of the impedance by weighting coefficients and then summing them.

10. The method according to claim 9, characterized in that, When the resonant frequency offset function value is large, the spatial receptive field parameter is increased to expand the range of interest; when the resonant frequency offset function value is small, the spatial receptive field parameter is decreased to focus on local detailed features.

Images