Efficient grounding electrode resistance reduction performance optimization design method based on multi-material composite

Abstract

The application provides a high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material compounding, relates to the technical field of electric power engineering, and comprises the following steps: extracting soil characteristics through a deep neural network, analyzing parameter correlation in combination with a transformer technology, determining grounding electrode parameters by applying a particle swarm algorithm, optimizing material proportions by using a deep deterministic policy gradient algorithm, and outputting a design scheme with optimal resistance reduction performance, so that the equivalent resistivity can be effectively reduced, the efficiency of a grounding system is improved, the complex soil environment is adapted, and the safety risk of an electric power system is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power engineering, and in particular to a high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material composite. BACKGROUND

[0002] With the wide application of power systems and electronic communication equipment, the grounding system is a key component for ensuring the safe operation of equipment and personal safety, and its performance directly affects the safe and stable operation of the entire system. Traditional grounding electrode systems are usually constructed with a single material, which is difficult to obtain an ideal grounding resistance value in a complex and variable soil environment, especially in high-resistivity soil areas. Conventional grounding methods often fail to meet the safety standard requirements. Therefore, multi-material composite grounding technology has gradually become an effective solution to reduce grounding resistance. By filling specific conductive materials such as graphite, activated carbon, and conductive concrete around the grounding electrode, a composite grounding system is formed.

[0003] However, the existing technology still lacks accurate modeling capability for the complex relationship between soil environmental parameters and grounding performance, cannot accurately predict the actual resistance reduction effect under variable soil conditions, resulting in a large deviation between the design scheme and the actual effect. Material proportioning design usually uses fixed proportions or trial-and-error methods, lacks systematic optimization means, and is difficult to find the best material combination scheme for specific soil conditions, causing resource waste and performance instability. There is a lack of a comprehensive consideration of the balance mechanism between resistance reduction effect and design complexity, excessive pursuit of resistance reduction effect while ignoring implementation difficulty and economic factors, and low practical application feasibility despite significant theoretical resistance reduction effect. SUMMARY

[0004] The present application provides a high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material composite, which can at least solve some of the problems in the prior art.

[0005] In a first aspect, the present application provides a high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material composite, comprising:

[0006] Obtain the soil environmental parameters of the grounding electrode installation area and perform feature extraction through a deep neural network to generate a soil feature vector. Calculate the correlation between different types of data in the soil environmental parameters through a transformer encoder combined with a multi-head attention mechanism. Determine the initial prediction data of the resistance reduction performance based on the correlation. Based on the initial prediction data of the resistance reduction performance and the soil environmental feature vector, solve the grounding electrode optimization parameters through a particle swarm algorithm.

[0007] The ground electrode optimization parameters are taken as constraint conditions, a deep deterministic policy gradient algorithm is used to calculate material ratio parameters, a predicted equivalent resistivity value is minimized to set an objective function for dynamic adjustment, historical data in an experience replay buffer is used to optimize the material ratio parameters, and an actual equivalent resistivity value is calculated based on the optimized material ratio parameters;

[0008] A reduction percentage of the actual equivalent resistivity value relative to an optimal equivalent resistivity value in a historical optimization record is calculated, material ratio parameter ranges in a current design scheme are counted and design complexity is evaluated, a performance evaluation result is calculated by taking the reduction percentage and the design complexity as evaluation bases, if the performance evaluation result meets a preset performance target, the ground electrode optimization parameters, the actual equivalent resistivity value and the optimized material ratio parameters are taken as an optimal design scheme and output.

[0009] In an optional implementation,

[0010] Soil environment parameters of a ground electrode installation area are acquired and feature extraction is performed through a deep neural network, a soil feature vector is generated, and an association between different types of data in the soil environment parameters is calculated through a transformer encoder combined with a multi-head attention mechanism, including:

[0011] Soil environment parameters of a ground electrode installation area are acquired, soil resistivity data at different depths are collected through a Wenner four-electrode method, an apparent resistivity data matrix is calculated based on measured current, voltage values and electrode spacing, soil dielectric constant is measured through a time domain reflector probe and soil water content data are calculated, and geological layering data are acquired through vertical electrical sounding;

[0012] The apparent resistivity data matrix, the soil water content data and the geological layering data are standardized and reorganized into three-channel input data, convolution operation and pooling processing are performed on the three-channel input data to obtain a soil feature vector, position encoding information is generated based on data collection depth and sampling spacing and added to the soil feature vector, and the soil feature vector is divided into multiple feature groups, a transformer encoder is used to calculate a query matrix, a key matrix and a value matrix of each feature group, and an association between different types of data in the soil environment parameters is calculated through a multi-head attention mechanism.

[0013] In an optional implementation,

[0014] Based on the association, initial prediction data of resistance reduction performance are determined, and the initial prediction data of resistance reduction performance and the soil environment feature vector are used to solve the ground electrode optimization parameters through a particle swarm algorithm, including:

[0015] An association matrix is constructed based on the association relationship, a weight coefficient of each type of data in the soil environment parameter is calculated by a standard mutual information method, an exponential coupling function between different types of data is established in combination with soil physical properties, and initial prediction data of the resistance reduction performance is calculated in combination with the weight coefficient;

[0016] An optimization objective function is constructed based on the initial prediction data of the resistance reduction performance and the soil environment feature vector, positions and velocities of the particle swarm are initialized based on geometric parameters and material parameters corresponding to the grounding electrode, the optimization objective function is solved by an adaptive mutation particle swarm algorithm, a mutation probability is calculated according to the number of iterations to update the particle distribution, an adaptability value of each particle after updating is calculated, and the particles are divided into an exploration group and a development group based on the adaptability value, a migration probability between the exploration group and the development group is calculated, and the particle distribution is updated based on the migration probability;

[0017] A particle with the highest adaptability value in the updated particle distribution is selected for local search to obtain a local optimal solution, if the adaptability value of the local optimal solution is greater than that of the current global optimal solution, the global optimal position is updated, a mean square deviation of the adaptability value of the particle swarm and a Lyapunov function value are calculated, and when the mean square deviation is less than a preset mean square deviation threshold and the Lyapunov function value is monotonically decreasing, the global optimal position is decoded as the output of the grounding electrode optimization parameter.

[0018] In an optional implementation,

[0019] The material ratio parameters are calculated by using a deep deterministic policy gradient algorithm, the target function of minimizing the predicted equivalent resistivity value is set, the material ratio parameters are dynamically adjusted, and the material ratio parameters are optimized by using historical data in an experience replay buffer, and the actual equivalent resistivity value is calculated based on the optimized material ratio parameters, including:

[0020] The grounding electrode optimization parameters are obtained, a mass percentage vector of the material ratio parameters is established according to the grounding electrode optimization parameters, the material ratio feasible region is constructed by taking the grounding electrode optimization parameters as constraint conditions, and the sum of the mass percentage vector is a constant and the single material ratio is in a preset range as constraint conditions of the material ratio feasible region;

[0021] A mapping function of the material ratio parameters and the equivalent resistivity is constructed, the mass percentage vector is mapped to the equivalent resistivity prediction value by a rectified linear unit function based on a multi-layer perception in the mapping function, a state space is constructed based on the mass percentage vector and the environmental parameters, a material ratio adjustment amount is generated as an action space, and a reward function value is calculated according to the equivalent resistivity prediction value, a constraint condition violation degree, and a difference between the material ratios at adjacent time points;

[0022] The reward function value is adjusted to obtain an updated mass percentage vector, and the state space, action space, reward function value and updated mass percentage vector are stored in an experience replay buffer; based on historical data in the experience replay buffer, an optimized material ratio parameter is obtained through a policy gradient method and a time difference error; the optimized material ratio parameter is subjected to dynamic attenuation of exploration noise and soft update of parameters to obtain an optimal material ratio parameter; and the optimal material ratio parameter is combined with geometric parameters in the grounding electrode optimization parameter to calculate an actual equivalent resistivity value.

[0023] In an optional embodiment,

[0024] Based on historical data in the experience replay buffer, an optimized material ratio parameter is obtained through a policy gradient method and a time difference error; the optimized material ratio parameter is subjected to dynamic attenuation of exploration noise and soft update of parameters to obtain an optimal material ratio parameter, including:

[0025] Randomly sampling historical data from the experience replay buffer to obtain a transition four-tuple composed of a state space, an action space, a reward value and a next state space corresponding to the historical data; calculating a current time reward value based on the transition four-tuple and a pre-set reward weight coefficient; constructing a policy function based on the state space and the action space, and calculating a gradient value of the policy function with respect to the material ratio parameter; calculating a parameter update amount based on the gradient value and a pre-set learning rate, and updating the policy function parameter in combination with a pre-set action value function;

[0026] Based on the current time reward value and the action value function, a time difference error is calculated in combination with the next state space; the policy function parameter is updated based on the time difference error and the learning rate; and an optimized material ratio parameter is calculated based on the updated policy function parameter and the value function parameter;

[0027] The optimized material ratio parameter is superimposed with exploration noise to obtain a noisy material ratio parameter; the noisy material ratio parameter is subjected to dynamic attenuation; the dynamically attenuated material ratio parameter is subjected to soft update through a pre-set soft update coefficient to obtain an optimal material ratio parameter.

[0028] In an optional embodiment,

[0029] A percentage reduction of the actual equivalent resistivity value with respect to an optimal equivalent resistivity value in a historical optimization record is calculated; and a material ratio parameter range in a current design scheme is counted and a design complexity is evaluated, including:

[0030] obtaining an actual equivalent resistivity value in a current design scheme, reading an optimal equivalent resistivity value in a historical optimization record, calculating a difference between the optimal equivalent resistivity value and the actual equivalent resistivity value, and calculating a reduction percentage based on the difference;

[0031] extracting material proportioning parameters in the current design scheme, calculating a Shannon entropy corresponding to the material proportioning parameters, calculating an accumulated sum of absolute values of differences between two types of parameters in the material proportioning parameters, and calculating a Gini coefficient based on the accumulated sum and the parameter types;

[0032] calculating a parameter correlation matrix between the material proportioning parameters by calculating a covariance and a standard deviation, calculating a maximum eigenvalue of the parameter correlation matrix, and calculating a design complexity of the current design scheme based on the Shannon entropy, the Gini coefficient, and the maximum eigenvalue.

[0033] In an optional embodiment,

[0034] calculating a performance evaluation result based on the reduction percentage and the design complexity as evaluation bases, and outputting grounding electrode optimization parameters, an actual equivalent resistivity value, and optimized material proportioning parameters as an optimal design scheme if the performance evaluation result meets a preset performance target, including:

[0035] obtaining the reduction percentage and the design complexity for normalization respectively, constructing an evaluation factor set and an evaluation level set based on a normalization result, and constructing a fuzzy relationship matrix based on membership degrees of each factor in the evaluation factor set to each level in the evaluation level set;

[0036] calculating weight values of each factor in the evaluation factor set and constructing a factor weight vector by using an analytic hierarchy process, multiplying the factor weight vector and the fuzzy relationship matrix to obtain a comprehensive evaluation vector, multiplying each component in the comprehensive evaluation vector and a score of a corresponding level and summing to obtain a performance score, and calculating a performance target based on a historical highest performance score and a historical performance score standard deviation obtained in advance;

[0037] determining whether the current design scheme meets the performance target based on the performance score, and combining the grounding electrode optimization parameters, the actual equivalent resistivity value, and the optimized material proportioning parameters into an optimal design scheme and outputting if the current design scheme meets the performance target.

[0038] In a second aspect of the embodiment of the application, an electronic device is provided, including:

[0039] a processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0040] In a third aspect, the present application provides a computer readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the method described above.

[0041] In the present application, the soil environment parameters are extracted and analyzed by the deep neural network and the transformer encoder, the precise prediction of the grounding electrode resistance reduction performance is realized, the adaptability and accuracy of the design scheme are significantly improved, the deep deterministic policy gradient algorithm is used to optimize the material ratio parameters, the historical data of the experience replay buffer is used for dynamic adjustment, the equivalent resistivity of the grounding electrode is effectively reduced, the resistance reduction effect is improved, the design complexity and material cost are reduced, the practicability and economy of the design scheme are ensured through comprehensive evaluation of the equivalent resistivity reduction percentage and the design complexity, and reliable technical support is provided for the engineering application of the grounding electrode. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A flowchart of the efficient grounding electrode resistance reduction performance optimization design method based on multi-material composite of the embodiment of the present application is shown in

[0043] Figure 2 A material ratio parameter optimization flowchart of the efficient grounding electrode resistance reduction performance optimization design method based on multi-material composite of the embodiment of the present application is shown in DETAILED DESCRIPTION

[0044] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] The technical scheme of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described in some embodiments.

[0046] Figure 1 A flowchart of the efficient grounding electrode resistance reduction performance optimization design method based on multi-material composite of the embodiment of the present application is shown in Figure 1 As shown in

[0047] Soil environmental parameters of the grounding electrode installation area are obtained and features are extracted through a deep neural network to generate soil feature vectors. The correlation between different types of data in the soil environmental parameters is calculated by a transformer encoder combined with a multi-head attention mechanism. Based on the correlation, the initial prediction data of resistance reduction performance is determined. Based on the initial prediction data of resistance reduction performance and the soil environmental feature vectors, the grounding electrode optimization parameters are solved by particle swarm optimization.

[0048] Using the grounding electrode optimization parameters as constraints, the material ratio parameters are calculated using a deep deterministic strategy gradient algorithm. The predicted equivalent resistivity value is minimized as the objective function for dynamic adjustment. Historical data in the experience replay buffer is used to optimize the material ratio parameters. The actual equivalent resistivity value is calculated based on the optimized material ratio parameters.

[0049] Calculate the percentage reduction of the actual equivalent resistivity value relative to the optimal equivalent resistivity value in the historical optimization record, statistically analyze the range of material ratio parameters in the current design scheme and evaluate the design complexity, use the percentage reduction and design complexity as the evaluation basis to calculate the performance evaluation result, if the performance evaluation result meets the preset performance target, then output the optimized grounding electrode parameters, the actual equivalent resistivity value and the optimized material ratio parameters as the optimal design scheme.

[0050] In one alternative implementation,

[0051] Soil environmental parameters of the grounding electrode installation area are obtained, and features are extracted using a deep neural network to generate soil feature vectors. The correlation between different types of data in the soil environmental parameters is then calculated using a transformer encoder combined with a multi-head attention mechanism, including:

[0052] Soil environmental parameters of the grounding electrode installation area are obtained. Soil resistivity data are collected at different depths using the Wenner quadrupole method. The apparent resistivity data matrix is ​​calculated based on the measured current, voltage values ​​and electrode spacing. Soil dielectric constant is measured using a time-domain reflectometry probe and soil moisture content data is calculated. Geological stratification data is obtained through vertical electrical sounding.

[0053] The apparent resistivity data matrix, soil moisture content data, and geological stratification data are standardized and recombined into three-channel input data. Convolution and pooling operations are performed on the three-channel input data to obtain soil feature vectors. Location encoding information is generated based on the data acquisition depth and sampling interval and added to the soil feature vectors, which are then divided into multiple feature groups. The query matrix, key matrix, and value matrix of each feature group are calculated using a transformer encoder. The correlation between different types of data in soil environmental parameters is calculated using a multi-head attention mechanism.

[0054] Soil resistivity data were collected at different depths within the grounding electrode installation area using the Wenner four-electrode method. Four electrodes were evenly spaced along a straight line in the target area, with the two outer electrodes serving as current electrodes and the two inner electrodes as potential electrodes. The spacing between the current electrodes was set to 20 meters, and the spacing between the potential electrodes was 5 meters. An alternating current (AC) signal with a frequency of 97 Hz and a current intensity of 200 mA was applied to the current electrodes, and the voltage difference between the potential electrodes was measured. For depth sampling, measurements were taken every 0.5 meters from the surface until a depth of 10 meters, resulting in 20 depth points. The apparent resistivity value was obtained by substituting the measured current value (I), voltage value (V), and electrode spacing (a) into the calculation formula. For example, at a depth of 2 meters, if the measured current I is 200mA, the voltage V is 0.85V, and the electrode spacing a is 5 meters, the apparent resistivity value is 133.52 ohm-meters. Data collected from multiple measuring points at different depths will eventually form an apparent resistivity data matrix with dimensions m×n, where m represents the number of depth sampling points and n represents the number of horizontal sampling points.

[0055] Soil dielectric constant was measured using a time-domain reflectometry (TDRS) probe. The probe was inserted into the soil at different depths, emitting electromagnetic pulses and receiving the reflected signals. After measuring the soil dielectric constant ε, the soil moisture content was calculated based on the relationship between the dielectric constant and soil moisture content. For example, when the measured soil dielectric constant was 12.5, the corresponding volumetric water content was approximately 18.7%. A total of 200 soil moisture content data points were collected throughout the target area at 0.5-meter depth intervals and 5-meter horizontal intervals.

[0056] Geological stratification data is obtained through vertical electrical sounding technology. Electrodes are deployed in the target area to measure the changes in resistivity at different depths, identify the interface locations and characteristics of different geological layers, and, based on the measurement results, divide the geological structure into layers such as topsoil (0-1.5 meters, resistivity of about 150 ohms·m), silty clay layer (1.5-4 meters, resistivity of about 80 ohms·m), sandy soil layer (4-7 meters, resistivity of about 200 ohms·m), and bedrock layer (below 7 meters, resistivity of about 500 ohms·m).

[0057] The collected raw data were standardized to ensure that the mean of each data type was 0 and the standard deviation was 1. The apparent resistivity data matrix, soil moisture content data, and geological stratification data were then reorganized into a three-channel input data structure, similar to that of an RGB image. The data matrix has a dimension of m×n×3, where the first channel is the apparent resistivity data, the second channel is the soil moisture content data, and the third channel is the geological stratification data.

[0058] Convolution and pooling operations are performed on the three-channel input data to extract soil features. Two convolutional layers are used for feature extraction: the first layer has a 3×3 kernel size and a stride of 1, using 16 kernels; the second layer has a 3×3 kernel size and a stride of 1, using 32 kernels. After each convolutional layer, the ReLU activation function is used to increase non-linearity, and max pooling is applied to reduce feature dimensionality. The pooling window size is 2×2, and the stride is 2. This process converts the original three-channel data into a soil feature vector, extracting key features from apparent resistivity, water content, and geological stratification data.

[0059] Location coding information is generated based on the data acquisition depth and sampling interval. For a sampling point at a depth of d meters and a horizontal position of h meters, the location coding is generated by a combination of sine and cosine functions, converting the location information into a representation in a high-dimensional vector space. The generated location coding information is added to the soil feature vector, and the feature vector with added location coding is divided into multiple feature groups, each containing features from adjacent depth intervals. For example, data from 0-2 meters, 2-4 meters, 4-6 meters, 6-8 meters, and 8-10 meters can be divided into five feature groups respectively.

[0060] The feature groups are processed using a transformer encoder. For each feature group, a query matrix Q, a key matrix K, and a value matrix V are calculated. The number of attention heads is set to 8, and each attention head has a dimension of 64. A multi-head attention mechanism is used to calculate the correlation between different feature groups, capturing the relationships between apparent resistivity, soil moisture content, and geological stratification.

[0061] In this embodiment, deep feature information of soil resistivity, water content and geological stratification is effectively extracted through convolution and pooling operations. Then, spatial depth information is introduced into feature representation by combining location encoding, so that the spatial distribution features of the soil environment are fully preserved. By using the transformer encoder and multi-head attention mechanism, the intrinsic relationship between different types of soil parameters can be modeled, realizing the deep fusion of multi-source information and the mining of correlation relationships. This not only improves the accuracy and reliability of soil environmental parameter analysis, but also provides a more scientific basis for soil characteristic assessment for grounding electrode installation.

[0062] In one alternative implementation,

[0063] Based on the correlation, initial prediction data for resistivity reduction performance is determined. Based on this initial prediction data and the soil environmental feature vector, the optimal parameters for the grounding electrode are solved using a particle swarm optimization algorithm, including:

[0064] Based on the aforementioned correlation, an association matrix is ​​constructed. The weight coefficient of each type of data in the soil environmental parameters is calculated using the standardized mutual information method. An exponential coupling function between different types of data is established in combination with soil physical properties, and the initial prediction data of drag reduction performance is obtained by combining the weight coefficient.

[0065] An optimization objective function is constructed based on the initial prediction data of drag reduction performance and the soil environment feature vector. The position and velocity of the particle swarm are initialized based on the geometric and material parameters corresponding to the grounding electrode. The optimization objective function is solved by an adaptive mutation particle swarm algorithm. The mutation probability is calculated based on the number of iterations to update the particle position. The fitness value of each particle after the update is calculated and the particles are divided into an exploration group and a development group based on the fitness value. The migration probability between the exploration group and the development group is calculated and the particle distribution is updated based on the migration probability.

[0066] The particle with the highest fitness value in the updated particle distribution is selected for local search to obtain a local optimum. If the fitness value of the local optimum is greater than the fitness value of the current global optimum, the global optimum position is updated. The mean square error of the particle swarm fitness value and the Lyapunov function value are calculated. When the mean square error is less than the preset mean square error threshold and the Lyapunov function value is monotonically decreasing, the global optimum position is decoded into grounding electrode optimization parameters for output.

[0067] An association matrix R is constructed using the association relationships obtained from the transformer encoder. The matrix has a dimension of 3×3 and represents the association strength between the three types of data: apparent resistivity, soil moisture content, and geological stratification. Each element Rij in the matrix represents the degree of association between the i-th type of data and the j-th type of data, with a value range of [0, 1]. The larger the value, the stronger the association. For example, for a certain test area, the association strength between apparent resistivity and soil moisture content in the association matrix is ​​0.83, the association strength between apparent resistivity and geological stratification is 0.67, and the association strength between soil moisture content and geological stratification is 0.58.

[0068] The weight coefficients of the three types of soil environmental parameters were calculated using the standardized mutual information method. The correlation strength between each type of data and the other two types of data was normalized to obtain the weight coefficients W1, W2 and W3, which represent the weights of apparent resistivity, soil moisture content and geological stratification data, respectively. Taking the above correlation matrix as an example, the weight coefficient W1 of apparent resistivity was calculated to be 0.42, the weight coefficient W2 of soil moisture content was 0.36 and the weight coefficient W3 of geological stratification data was 0.22.

[0069] Based on soil physical properties, an exponential coupling function is established between different types of data. This function considers the nonlinear variation of soil resistivity with water content and the influence of geological stratification on current distribution. Taking soil water content μ as the independent variable and resistivity ρ as the dependent variable, as an example, when soil water content increases from 5% to 30%, resistivity decreases from 500 ohm-meters to 50 ohm-meters, exhibiting an exponential decay characteristic. Combining the weighting coefficients and the exponential coupling function, the initial predicted data P for resistivity reduction performance is calculated. Taking the target area as an example, when the apparent resistivity is 120 ohm-meters, the soil water content is 18%, and the geological structure is three-layered, the predicted grounding electrode resistivity reduction performance P is 42 ohms.

[0070] Based on the initial predicted data P of the grounding resistance reduction performance and the soil environmental feature vector F, an optimization objective function J is constructed. The optimization objective function comprehensively considers factors such as grounding resistance value, material cost, and installation difficulty, with the goal of minimizing the grounding resistance while controlling the cost within a reasonable range. The objective function is in the form of J(x), where x is the parameter vector to be optimized, including the geometric parameters of the grounding electrode (length, diameter, burial depth, etc.) and material parameters (material type, ratio, etc.).

[0071] Initialize the position and velocity of the particle swarm to solve the objective function. Set the particle swarm size to 50, with each particle containing 10 dimensions, corresponding to 10 parameters to be optimized. Parameters include grounding electrode length (range 2-6 meters), diameter (range 10-50 millimeters), burial depth (range 0.5-3 meters), graphite content (range 0-40%), bentonite content (range 0-30%), activated carbon content (range 0-20%), etc. The initial particle position is randomly generated within the parameter space, and the velocity is initialized to 5% of the corresponding parameter range. For example, the initial position of the first particle is [3.5 meters, 25 millimeters, 1.8 meters, 15%, 10%, 5%], and the initial velocity is [0.2 meters, 2 millimeters, 0.1 meters, 2%, 1%, 1%].

[0072] The objective function is solved using an adaptive mutation particle swarm optimization algorithm. The mutation probability Pm is calculated based on the iteration number t. Initially, Pm is relatively large (approximately 0.3), gradually decreasing as iterations progress (down to around 0.05). The mutation operation is achieved by adding a random perturbation to a certain dimension of the particle, with the perturbation magnitude inversely proportional to the current iteration number. For example, in the 10th iteration, a random perturbation of ±0.2 meters is added to the length parameter of a particle.

[0073] The fitness value of each particle after the update is calculated. The fitness value is equal to the reciprocal of the objective function J(x). The higher the fitness value, the better the performance of the grounding electrode design scheme represented by that particle. Based on the fitness value, the particle swarm is divided into an explorer group and a development group. The top 30% of particles in terms of fitness value are assigned to the development group, responsible for local fine-grained search; the remaining particles are assigned to the explorer group, responsible for global search. The migration probability Pt between the explorer group and the development group is calculated. The migration probability is proportional to the difference in fitness between the two groups. For example, when the average fitness of the development group is 50% higher than that of the explorer group, the migration probability is set to 0.15.

[0074] Updating particle distribution based on migration probability allows some particles from the exploration group to migrate to the vicinity of the development group, enhancing the algorithm's global convergence capability. For example, particles with relatively high fitness in the exploration group move towards the direction of the optimal particle in the development group with a probability of 0.15, updating their positions.

[0075] The particle with the highest fitness value is selected from the updated particle distribution for local search. A simulated annealing strategy is then used to perform a fine-grained search within the particle's neighborhood to obtain a local optimum. For example, the grounding electrode parameters corresponding to the particle with the highest fitness are [4.2 m, 30 mm, 2.1 m, 25%, 15%, 8%]. Through local search, these parameters are adjusted to [4.25 m, 32 mm, 2.05 m, 26%, 15.5%, 7.8%], resulting in a 5% increase in fitness value.

[0076] The fitness value of a local optimum is checked against the fitness value of the current global optimum. If the fitness value is greater, the global optimum position is updated. The global optimum position and its fitness value are recorded for each iteration, forming the trajectory of the optimization process. For example, in the 20th iteration, the fitness value of the global optimum is 0.024, corresponding to a grounding resistance of 41.7 ohms; in the 50th iteration, it increases to 0.028, corresponding to a grounding resistance of 35.8 ohms.

[0077] The mean squared error (MSE) of the particle swarm fitness values ​​and the Lyapunov function value are calculated to determine whether the algorithm has converged. The MSE reflects the degree of aggregation of the particle distribution, while the Lyapunov function value reflects the stability of the system. When the MSE of the particle swarm fitness values ​​is less than a preset MSE threshold (e.g., 0.001) and the Lyapunov function value monotonically decreases, the algorithm is considered converged, and the global optimal position is decoded into optimized grounding electrode parameters. The final optimized parameters are: grounding electrode length 4.5 meters, diameter 35 millimeters, burial depth 2.2 meters, graphite content 28%, bentonite content 16%, activated carbon content 8%, etc.

[0078] In this embodiment, an association matrix is ​​constructed and weighted processing of different types of soil data is performed using the standardized mutual information method. Then, the multi-source soil parameters are fused and modeled by combining the exponential coupling function, so that the predicted results of the resistance reduction performance are more consistent with the actual soil physical characteristics. The adaptive mutant particle swarm optimization algorithm is used to solve the optimization objective function. By co-evolution and migration probability adjustment of the exploratory and development groups, the particle swarm is effectively prevented from getting trapped in local optima, and high-precision optimization of the grounding electrode parameters is achieved. This makes the optimization results have better global convergence and stability, and significantly improves the resistance reduction performance of the grounding electrode.

[0079] In one alternative implementation,

[0080] Using the grounding electrode optimization parameters as constraints, a deep deterministic gradient algorithm is employed to calculate the material mix parameters. Minimizing the predicted equivalent resistivity is set as the objective function for dynamic adjustment. Historical data from the experience replay buffer is used to optimize the material mix parameters. Based on the optimized material mix parameters, the actual equivalent resistivity values ​​are calculated, including:

[0081] Obtain the grounding electrode optimization parameters, establish a mass percentage vector of material proportion parameters based on the grounding electrode optimization parameters, construct the material proportion feasible region with the grounding electrode optimization parameters as constraints, and take the sum of the mass percentage vectors as a constant value and the individual material proportions within a preset range as constraints of the material proportion feasible region.

[0082] A mapping function between material ratio parameters and equivalent resistivity is constructed. The mass percentage vector is mapped to the predicted equivalent resistivity value through the rectified linear unit function based on multilayer perceptron in the mapping function. A state space is constructed based on the mass percentage vector and environmental parameters, and the material ratio adjustment amount is generated as the action space. The reward function value is calculated based on the predicted equivalent resistivity value, the degree of constraint violation, and the difference in material ratio between adjacent time points.

[0083] The updated mass percentage vector is obtained by adjusting the material ratio parameters according to the reward function value. The state space, action space, reward function value, and updated mass percentage vector are stored in the experience replay buffer. Based on the historical data in the experience replay buffer, the optimized material ratio parameters are obtained through the policy gradient method and time-series differential error. The optimized material ratio parameters are subjected to dynamic noise attenuation and soft parameter update to obtain the optimal material ratio parameters. The actual equivalent resistivity value is calculated by combining the geometric parameters in the grounding electrode optimization parameters.

[0084] The optimized parameters of the grounding electrode are obtained, including a grounding electrode length of 4.5 meters, a diameter of 35 millimeters, a burial depth of 2.2 meters, a graphite content of 28%, a bentonite content of 16%, and an activated carbon content of 8%. Based on the material-related parameters in the optimized parameters of the grounding electrode, a mass percentage vector of material proportion parameters is established. The mass percentage vector contains the mass percentage of various materials in the composite grounding electrode. In this embodiment, the mass percentage vector contains six components: graphite powder (28%), bentonite (16%), activated carbon (8%), conductive polymer (5%), cement (36%), and other additives (7%).

[0085] A feasible region for material proportioning is constructed using the grounding electrode optimization parameters as constraints. The feasible region refers to the set of material proportioning parameters that satisfy all constraints, including: the sum of the mass percentage vectors M must equal 100%; and the proportions of individual materials must be within preset ranges. Specifically, these preset ranges are: graphite powder content 20%-35%, bentonite content 10%-25%, activated carbon content 5%-15%, conductive polymer content 2%-10%, cement content 30%-45%, and other additive content 3%-10%. These constraints ensure the physical feasibility and engineering practicality of the material proportioning.

[0086] A mapping function between material proportioning parameters and equivalent resistivity is constructed. This function maps the mass percentage vector to a predicted equivalent resistivity value. The mapping function employs a multilayer perceptron structure, comprising an input layer, two hidden layers, and an output layer. The input layer has 6 nodes, corresponding to the 6 components of the mass percentage vector; the first hidden layer has 12 nodes, the second hidden layer has 8 nodes, and the output layer has 1 node corresponding to the predicted equivalent resistivity value. The hidden layers use Rectified Linear Units (ReLU) as the activation function to ensure the network has nonlinear expressive capabilities. For example, when the input mass percentage vector is [28%, 16%, 8%, 5%, 36%, 7%], the predicted equivalent resistivity value calculated by the mapping function is 0.25 ohm-meters.

[0087] A state space is constructed based on the mass percentage vector and environmental parameters. This state space, representing the state in the deep reinforcement learning algorithm, has 12 dimensions and includes the six components of the mass percentage vector and six key environmental parameters (average soil resistivity, water content, pH value, temperature, ion concentration, and number of geological strata). The resulting material ratio adjustment amounts serve as the action space, which has 6 dimensions and corresponds to the adjustment amounts of the six components of the mass percentage vector, with the adjustment range limited to ±2%.

[0088] The reward function value is calculated based on the predicted equivalent resistivity, the degree of constraint violation, and the difference in material ratios between adjacent time points. The reward function consists of three parts: equivalent resistivity reward, constraint reward, and smoothness reward. The equivalent resistivity reward is inversely proportional to the predicted equivalent resistivity; the lower the equivalent resistivity, the higher the reward. The constraint reward is inversely proportional to the degree of constraint violation; the smaller the violation, the higher the reward. The smoothness reward is inversely proportional to the difference in material ratios between adjacent time points; the smaller the difference, the higher the reward. For the aforementioned material ratios, assuming the calculated equivalent resistivity reward is 15, the constraint reward is 10 (all constraints are satisfied), and the smoothness reward is 8, the total reward function value is 33.

[0089] The material ratio parameters are adjusted based on the reward function value to obtain the updated mass percentage vector. The adjustment is performed using a policy network, which takes the state space as input and outputs the adjustment amount in the action space. For example, for the original mass percentage vector [28%, 16%, 8%, 5%, 36%, 7%], the policy network outputs the adjustment amount as [+1%, -0.5%, +0.5%, 0%, -1%, 0%], and the adjusted mass percentage vector is [29%, 15.5%, 8.5%, 5%, 35%, 7%].

[0090] The adjustments in the state space and action space, the reward function value, and the updated quality percentage vector are stored in the experience replay buffer. The experience replay buffer has a capacity of 1000 and manages data using a first-in, first-out (FIFO) approach. When the buffer reaches its capacity limit, new data replaces the oldest data stored.

[0091] Based on historical data in the experience replay buffer, optimized material ratio parameters are obtained through the policy gradient method and temporal difference error. The policy gradient method updates network parameters by calculating the gradient of the policy network π, while the temporal difference error is used to evaluate state value, improving learning efficiency. Specifically, an Actor-Critic architecture is adopted, where the Actor network is responsible for generating actions (material ratio adjustments), and the Critic network is responsible for evaluating state value. The training process uses a mini-batch learning method, randomly selecting 64 samples from the experience replay buffer for each training iteration. The learning rate is set to 0.001, and the discount factor is set to 0.95.

[0092] The optimized material ratio parameters were subjected to dynamic decay of exploration noise and soft parameter updates. Exploration noise refers to random perturbations added to the action space to promote the algorithm's exploration of a broader solution space. As the number of training iterations increased, the exploration noise linearly decreased from an initial value of 0.2 to 0.05, ensuring sufficient exploration capability in the early stages and focusing more on utilizing known superior strategies in later stages. Soft parameter updates refer to the target network parameters approaching the current network parameters at a small update rate (e.g., 0.01), enhancing the algorithm's stability. After 1000 iterations of optimization, the optimal material ratio parameters were obtained as follows: graphite powder (30%), bentonite (14%), activated carbon (10%), conductive polymer (6%), cement (33%), and other additives (7%).

[0093] Combining the geometric parameters (4.5 meters in length, 35 millimeters in diameter, and 2.2 meters in burial depth) and the optimal material ratio parameters in the grounding electrode optimization parameters, the actual equivalent resistivity value is calculated to be 0.18 ohm-meters.

[0094] In this embodiment, by establishing a mapping relationship between material ratio parameters and equivalent resistivity, and combining policy gradient and temporal difference methods in reinforcement learning to dynamically optimize the material ratio, it is possible not only to automatically search for the optimal ratio scheme under the premise of meeting the constraints, but also to improve the stability and convergence efficiency of the model through experience replay and parameter soft update, thus realizing intelligent optimization of material ratio parameters, making the calculated actual equivalent resistivity lower, and effectively improving the resistance reduction effect and overall performance of the grounding electrode.

[0095] In one alternative implementation,

[0096] Based on historical data in the experience replay buffer, optimized material ratio parameters are obtained through the policy gradient method and temporal difference error. The optimized material ratio parameters are then subjected to dynamic noise attenuation and soft parameter updates to obtain the optimal material ratio parameters, including:

[0097] Historical data is randomly sampled from the experience replay buffer to obtain a transition quadruple consisting of the state space, action space, reward value, and next state space corresponding to the historical data. The reward value at the current moment is calculated based on the transition quadruple and the pre-set reward weight coefficient. A policy function is constructed based on the state space and the action space. The gradient value of the policy function with respect to the material ratio parameter is calculated. Based on the gradient value and the pre-set learning rate, the parameter update amount is calculated in combination with the preset action value function, and the policy function parameters are updated.

[0098] Based on the current reward value and the action value function, the temporal difference error is calculated in the next state space. The value function parameters are updated according to the temporal difference error and the learning rate. The optimized material ratio parameters are calculated based on the updated policy function parameters and value function parameters.

[0099] The optimized material ratio parameters are superimposed with exploration noise to obtain noisy material ratio parameters. The noisy material ratio parameters are dynamically attenuated, and the dynamically attenuated material ratio parameters are softly updated using a preset soft update coefficient to obtain the optimal material ratio parameters.

[0100] Historical data is randomly sampled from the experience replay buffer, with a batch size of 64, to obtain a transition quadruple consisting of the state space, action space, reward value, and next state space corresponding to the historical data. The state space contains material ratio parameters and environmental parameters, with a dimension of 12; the action space contains material ratio adjustment, with a dimension of 6; the reward value is a single scalar; and the next state space has the same structure as the state space, also a 12-dimensional vector. For example, the data for a certain transition quadruple is as follows: the state space is [29%, 15.5%, 8.5%, 5%, 35%, 7%, 125, 18%, 6.8, 22, 0.015, 3], corresponding to [graphite powder, bentonite, activated carbon, conductive polymer, cement, other additives, average soil resistivity, water content, pH value, temperature, ion concentration, number of geological strata]; the action space is [+1%, -0.5%, +0.5%, +1%, -2%, 0%], representing the adjustment amount of each material ratio; the reward value is 35; the next state space is [30%, 15%, 9%, 6%, 33%, 7%, 125, 18%, 6.8, 22, 0.015, 3].

[0101] The reward value at the current moment is calculated based on the transfer quadruple and pre-set reward weight coefficients. The reward weight coefficients are used to balance the importance of different reward components and are set as follows: equivalent resistivity reward weight is 0.6, constraint condition reward weight is 0.3, and smoothness reward weight is 0.1. For the aforementioned example, the equivalent resistivity reward is 40 (equivalent resistivity is reduced to 0.17 ohm-meters), the constraint condition reward is 25 (all constraints are satisfied), and the smoothness reward is 20 (material ratio changes are small). The comprehensive calculation yields a reward value of 0.6×40+0.3×25+0.1×20=35.5 at the current moment.

[0102] A policy function is constructed based on the state space and action space, implemented using a deep neural network. It comprises three layers: an input layer with 12 nodes, corresponding to the state space dimension; a hidden layer with 24 nodes, using the hyperbolic tangent function as the activation function; and an output layer with 6 nodes, corresponding to the action space dimension, using the sigmoid function to normalize the output values ​​and map them to the range [-2%, +2%]. The policy function's parameters include the weight matrices and bias vectors between layers, initially randomly generated, for a total of approximately 480 parameters.

[0103] The gradient of the policy function with respect to the material proportion parameters is calculated using the backpropagation algorithm. The gradient of the loss function with respect to the output layer is also calculated, and this process is repeated layer by layer backpropagating to the input layer. The loss function is designed to be the negative of the reward value at the current moment, maximizing the reward value during the optimization process. For the state space [29%, 15.5%, 8.5%, 5%, 35%, 7%, 125, 18%, 6.8, 22, 0.015, 3] in the aforementioned example, the gradient matrix shows that the gradient value for graphite powder is 0.028, the gradient value for bentonite is -0.015, the gradient value for activated carbon is 0.022, the gradient value for conductive polymer is 0.031, the gradient value for cement is -0.046, and the gradient value for other additives is 0.003.

[0104] Based on the gradient value and a pre-set learning rate, the parameter update amount is calculated and the policy function parameters are updated using a preset action value function. The learning rate is set to 0.001. The action value function uses a deep neural network with a similar structure to the policy function, but the output layer has only one node, representing the value estimate of selecting a specific action in the current state. The parameter update amount is calculated as the product of the learning rate, the gradient value, and the action value function. For graphite powder in the previous example, the parameter update amount is 0.001 × 0.028 × 42 = 0.001176, where 42 is the output value of the action value function. The updated policy function parameters are obtained by adding the parameter update amount.

[0105] The temporal difference error is calculated based on the current reward value and action value function, combined with the next state space. The temporal difference error reflects the difference between the current value estimate and the next state value estimate plus the reward. For the example data mentioned above, the action value function output for the current state is 42, the action value function output for the next state is 45, the current reward value is 35.5, and the discount factor is set to 0.95. Therefore, the temporal difference error is calculated as 35.5 + 0.95 × 45 - 42 = 36.25.

[0106] The value function parameters are updated based on the temporal difference error and the learning rate. The update direction of the value function parameters is proportional to the temporal difference error, and the update amount is the product of the learning rate and the temporal difference error. For the temporal difference error of 36.25 in the previous example, the update amount of the value function parameters is 0.001 × 36.25 = 0.03625. Using the gradient descent method, the update amount is applied to the parameters of each layer of the value function network to obtain the updated value function parameters.

[0107] The optimized material ratio parameters are calculated based on the updated policy function parameters and value function parameters. The current state space is input into the updated policy function to obtain the optimized material ratio adjustment amount. The adjustment amount is applied to the current material ratio parameters to obtain the optimized material ratio parameters. For the material ratio [29%, 15.5%, 8.5%, 5%, 35%, 7%] in the previous example, the optimized adjustment amount is [+1.2%, -0.7%, +0.8%, +1.1%, -2.4%, +0%], and the optimized material ratio parameters are [30.2%, 14.8%, 9.3%, 6.1%, 32.6%, 7%].

[0108] Noisy material proportion parameters are obtained by superimposing exploration noise on the optimized material proportion parameters. The exploration noise is generated using a normal distribution with a mean of 0 and an initial standard deviation of 0.2. For the optimized material proportion parameters [30.2%, 14.8%, 9.3%, 6.1%, 32.6%, 7%], the noisy material proportion parameters [30.5%, 14.6%, 9.5%, 6.3%, 32.1%, 7%] are obtained by superimposing exploration noise.

[0109] The material proportions with noise were dynamically attenuated. Dynamic attenuation refers to gradually reducing the standard deviation of the exploration noise as the number of training iterations increases, linearly decreasing from an initial value of 0.2 to a final value of 0.05. Assuming the current training is at the 500th iteration (out of a total of 1000 iterations), the current standard deviation of the exploration noise is 0.125. The material proportions with noise were dynamically attenuated, resulting in the dynamically attenuated material proportions [30.35%, 14.7%, 9.4%, 6.2%, 32.35%, 7%].

[0110] The optimal material ratio parameters are obtained by soft updating the dynamically decayed material ratio parameters using a preset soft update coefficient. The soft update coefficient is set to 0.1, which means that 10% of the new parameters and 90% of the old parameters are weighted averaged. For the dynamically decayed material ratio parameters [30.35%, 14.7%, 9.4%, 6.2%, 32.35%, 7%] and the original parameters [29%, 15.5%, 8.5%, 5%, 35%, 7%], the optimal material ratio parameters [29.135%, 15.42%, 8.59%, 5.12%, 34.735%, 7%] are obtained after soft updating.

[0111] After multiple iterations (usually 1000-2000 times), the material proportions gradually converge to the optimal solution. The optimal material proportions obtained are [30.5%, 14.2%, 9.8%, 6.5%, 32%, 7%], with a corresponding equivalent resistivity of 0.15 ohm-meters.

[0112] In this embodiment, by introducing an empirical replay sampling and transfer quadruples construction mechanism, and combining policy function gradient update and temporal difference error correction, not only can the accuracy of the policy and value function be improved during the material ratio optimization process, but the convergence ability under complex constraints can also be enhanced. By exploring the dynamic attenuation of noise and the soft update method, the relationship between global exploration and local convergence is effectively balanced, realizing efficient adaptive optimization of material ratio parameters, making the obtained optimal material ratio parameters more stable and reliable, and further improving the resistance reduction performance of the grounding electrode.

[0113] Figure 2 This is a flowchart illustrating the material ratio parameter optimization of the high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material composite, as described in an embodiment of the present invention.

[0114] In one alternative implementation,

[0115] Calculate the percentage reduction in actual equivalent resistivity relative to the optimal equivalent resistivity in historical optimization records, and statistically analyze the range of material proportioning parameters in the current design scheme and assess design complexity, including:

[0116] Obtain the actual equivalent resistivity value in the current design scheme, read the optimal equivalent resistivity value from the historical optimization record, calculate the difference between the optimal equivalent resistivity value and the actual equivalent resistivity value, and calculate the reduction percentage based on the difference.

[0117] Extract the material ratio parameters from the current design scheme, calculate the Shannon entropy corresponding to the material ratio parameters, calculate the sum of the absolute values ​​of the differences between each pair of different types of parameters in the material ratio parameters, and calculate the Gini coefficient by combining the sum and the parameter type.

[0118] The material proportioning parameters are calculated, and a parameter correlation matrix is ​​constructed by calculating the covariance and standard deviation. The eigenvalues ​​of the parameter correlation matrix are calculated to obtain the maximum eigenvalue. The design complexity of the current design scheme is calculated based on the Shannon entropy, the Gini coefficient, and the maximum eigenvalue.

[0119] The actual equivalent resistivity value of the current design scheme is obtained. In the aforementioned embodiment, the actual equivalent resistivity value of the current design scheme is 0.15 ohm-meters. The optimal equivalent resistivity value in the historical optimization record is read. The historical optimization record is stored in the optimization database and contains the equivalent resistivity values ​​of all past design schemes. In this embodiment, the optimal equivalent resistivity value in the historical record is 0.25 ohm-meters. This value comes from the traditional graphite-based grounding electrode formula. The difference between the optimal equivalent resistivity value and the actual equivalent resistivity value is calculated. The difference is 0.25 - 0.15 = 0.10 ohm-meters. The reduction percentage is calculated based on the difference. The reduction percentage is the difference divided by the historical optimal equivalent resistivity value and then multiplied by 100%, i.e., 0.10 ÷ 0.25 × 100% = 40%. This indicates that the equivalent resistivity of the current design scheme is reduced by 40% compared to the historical optimal scheme, and the resistance reduction effect is significantly improved.

[0120] The material proportions in the current design scheme are extracted, including graphite powder (30.5%), bentonite (14.2%), activated carbon (9.8%), conductive polymer (6.5%), cement (32%), and other additives (7%). The Shannon entropy corresponding to these material proportions is calculated. Shannon entropy is an indicator in information theory used to measure uncertainty; in this embodiment, it measures the uniformity of the material proportions. The calculation process for Shannon entropy is as follows: divide each material proportion percentage by 100 to obtain a probability value, and calculate the negative of the sum of the products of each probability value and its natural logarithm. For the current design scheme, the calculated Shannon entropy is 1.65. A higher Shannon entropy value indicates a more uniform distribution of material proportions; a lower Shannon entropy value indicates a more concentrated distribution of material proportions.

[0121] The absolute values ​​of the differences between each pair of parameters of different types in the material proportioning parameters are calculated. Six materials are paired, resulting in 15 possible combinations. The absolute value of the difference in proportion for each combination is calculated, and then all differences are summed. For example, the absolute value of the difference between graphite powder (30.5%) and bentonite (14.2%) is |30.5% - 14.2%| = 16.3%. The absolute values ​​of the differences for other combinations are calculated similarly, and the sum is 147.4%. The Gini coefficient is calculated by combining the sum with the number of parameter types. The Gini coefficient is an indicator in economics that measures the degree of uneven distribution. In this embodiment, it is used to measure the unevenness of the material proportioning. The sum is divided by the number of parameter types and multiplied by the maximum possible value of the sum. For six materials, the maximum possible value of the sum is 300%, therefore the Gini coefficient is 147.4% ÷ (6 × 300%) = 0.082. The closer the Gini coefficient is to 0, the more uniform the material ratio; the closer the Gini coefficient is to 1, the less uniform the material ratio.

[0122] The correlation between material proportioning parameters is calculated by constructing a parameter correlation matrix through covariance and standard deviation calculations. Based on material proportioning data from historical design schemes, the covariance between each material proportioning parameter is calculated. For example, to calculate the covariance between graphite powder and bentonite proportions, the historical trends in these two material proportions are needed. Assuming the historical data includes 20 design schemes, the calculated covariance between graphite powder and bentonite is -0.023, indicating a weak negative correlation between the two material proportions. The standard deviation of each material proportioning parameter is calculated; for example, the standard deviation for graphite powder is 0.032, and for bentonite, it is 0.018. The correlation coefficient is obtained by dividing the covariance by the product of the standard deviations of the two parameters. A 6×6 correlation matrix is ​​then constructed from the correlation coefficients of all material proportioning parameters.

[0123] The eigenvalues ​​of the parameter correlation matrix are calculated to obtain the largest eigenvalue. The eigenvalue calculation uses a power-law iteration method. A non-zero vector is randomly initialized, and matrix-vector multiplication is repeatedly performed, followed by normalization, until the vector converges. In this embodiment, after 50 iterations, the vector converges, and the largest eigenvalue of the parameter correlation matrix is ​​2.35. The larger the largest eigenvalue, the stronger the correlation between the material proportioning parameters, and the higher the design complexity.

[0124] The design complexity of the current design scheme is calculated based on Shannon entropy, Gini coefficient, and the largest eigenvalue. Design complexity is a comprehensive indicator of the overall complexity of a design scheme. It is calculated by multiplying Shannon entropy, Gini coefficient, and the largest eigenvalue by preset weighting coefficients and then summing them. The preset weighting coefficients are: Shannon entropy weight 0.3, Gini coefficient weight 0.3, and the largest eigenvalue weight 0.4. For the current design scheme, the design complexity is calculated as 0.3 × 1.65 + 0.3 × 0.082 + 0.4 × 2.35 = 1.44. Design complexity ranges from 0 to 5; higher complexity indicates a more complex design scheme and greater implementation difficulty. The current design scheme has a complexity of 1.44, which is relatively low.

[0125] In this embodiment, by comparing the actual equivalent resistivity with the historical best value and calculating the percentage reduction, the improvement effect of the current design scheme in resistance reduction performance can be intuitively quantified. Combined with the Shannon entropy, Gini coefficient, and maximum eigenvalue of the correlation matrix of the material ratio parameters, the complexity of the design scheme is comprehensively evaluated. This not only reflects the balance and differences in the material ratio, but also reveals the strength of the correlation between different parameters. It achieves a dual evaluation of the grounding electrode design scheme in terms of performance improvement and complexity, providing a scientific quantitative basis for design optimization.

[0126] In one alternative implementation,

[0127] The performance evaluation results are calculated based on the percentage reduction and design complexity. If the performance evaluation results meet the preset performance targets, the optimized grounding electrode parameters, actual equivalent resistivity values, and optimized material ratio parameters are output as the optimal design scheme, including:

[0128] The percentage reduction and design complexity are normalized respectively. Based on the normalization results, an evaluation factor set and an evaluation level set are constructed. Based on the membership degree of each factor in the evaluation factor set to each level in the evaluation level set, a fuzzy relation matrix is ​​constructed.

[0129] The weight values ​​of each factor in the evaluation factor set are calculated using the analytic hierarchy process and a factor weight vector is constructed. The factor weight vector is multiplied by the fuzzy relation matrix to obtain a comprehensive evaluation vector. Each component in the comprehensive evaluation vector is multiplied by the score of the corresponding level and summed to obtain a performance score. The performance target is calculated based on the pre-obtained highest historical performance score and the standard deviation of historical performance scores.

[0130] Based on the performance score, it is determined whether the current design scheme meets the performance target. If it does, the grounding electrode optimization parameters, the actual equivalent resistivity value, and the optimized material ratio parameters are combined into the optimal design scheme and output.

[0131] The calculated reduction percentage and design complexity were normalized. The reduction percentage was 40%, and the design complexity was 1.44. The normalization process transformed the original data to the interval [0, 1]. The normalization method used was the range method, which subtracted the historical minimum value from the original value and then divided by the historical range. Based on historical data, the minimum reduction percentage was 5%, and the maximum was 50%. Therefore, the normalized reduction percentage was (40% - 5%) / (50% - 5%) = 0.778. The historical minimum design complexity was 0.8, and the maximum was 4.2. Therefore, the normalized design complexity was (1.44 - 0.8) / (4.2 - 0.8) = 0.188. The normalized data is more suitable for comprehensive evaluation of multiple indicators.

[0132] Based on the normalization results, an evaluation factor set and an evaluation level set are constructed. The evaluation factor set includes two factors: reduction percentage and design complexity; the evaluation level set includes four levels: excellent, good, average, and poor. The membership degree of each evaluation factor to each evaluation level forms a fuzzy relation matrix. The membership degree represents the degree to which the factor belongs to a certain evaluation level when it takes a specific value, and the value range is [0, 1]. For the reduction percentage factor, the membership degree corresponding to the normalized value of 0.778 is: excellent level 0.85, good level 0.15, average level 0, and poor level 0. For the design complexity factor, the membership degree corresponding to the normalized value of 0.188 is: excellent level 0.9, good level 0.1, average level 0, and poor level 0. Combining these two sets of membership degree values, a fuzzy relation matrix is ​​constructed. The matrix dimension is 2×4, with each row representing an evaluation factor and each column representing an evaluation level.

[0133] The weight values ​​of each factor in the evaluation factor set are calculated using the analytic hierarchy process (AHP), and a factor weight vector and a judgment matrix are constructed. The relative importance of the percentage reduction and design complexity is determined based on pre-obtained expert experience. In this embodiment, the importance ratio of the percentage reduction to design complexity is 3, meaning the percentage reduction is 3 times more important than design complexity. Based on the judgment matrix, an eigenvector is calculated and normalized to obtain the weight vector. The calculation results are: a weight of 0.75 for the percentage reduction and a weight of 0.25 for the design complexity. These weight values ​​reflect the relative importance of each evaluation factor in the comprehensive evaluation. The higher weight of the percentage reduction indicates that, in grounding electrode design, the resistance reduction effect is more important than the design complexity.

[0134] The factor weight vector is multiplied by the fuzzy relation matrix to obtain the comprehensive evaluation vector. Each weight value in the weight vector is then multiplied by the element of the corresponding row in the fuzzy relation matrix, and the results are summed column-wise. In this embodiment, the percentage weight is reduced by 0.75 multiplied by its membership degree [0.85, 0.15, 0, 0], resulting in [0.6375, 0.1125, 0, 0]; the design complexity weight is reduced by 0.25 multiplied by its membership degree [0.9, 0.1, 0, 0], resulting in [0.225, 0.025, 0, 0]. These two values ​​are then summed column-wise to obtain the comprehensive evaluation vector [0.8625, 0.1375, 0, 0]. Each component in the comprehensive evaluation vector represents the comprehensive membership degree of the current design scheme to each evaluation level.

[0135] The performance score is obtained by multiplying each component of the comprehensive evaluation vector by its corresponding grade score and summing the results. The scores for each evaluation grade are set as follows: Excellent 95 points, Good 80 points, Average 65 points, and Poor 50 points. The performance score is calculated as 0.8625×95+0.1375×80+0×65+0×50=92.9375 points. The performance score is a quantitative evaluation of the comprehensive performance of the current design scheme; the higher the score, the better the scheme's performance.

[0136] Based on the pre-obtained highest historical performance score and the standard deviation of historical performance scores, the performance target is calculated. According to historical data, the highest historical performance score is 90 points, and the standard deviation of historical performance scores is 5 points. The performance target is set as the highest historical performance score plus half of the standard deviation, i.e., 90 + 5 × 0.5 = 92.5 points. The performance target is a threshold for judging whether the current design scheme meets the expected performance level; only when the performance score is higher than or equal to the performance target is the design scheme considered to meet the requirements.

[0137] The performance score determines whether the current design meets the performance target. The current design has a performance score of 92.9375, which is higher than the performance target of 92.5. Therefore, the current design meets the performance target requirements. After meeting the target, the optimized grounding electrode parameters, the actual equivalent resistivity value, and the optimized material ratio parameters are combined into the optimal design and output.

[0138] The details of the optimal design scheme include: optimized grounding electrode parameters (length 4.5 meters, diameter 35 mm, burial depth 2.2 meters), actual equivalent resistivity value (0.15 ohm-meters, 40% lower than the historical best value), and optimized material ratio parameters (graphite powder 30.5%, bentonite 14.2%, activated carbon 9.8%, conductive polymer 6.5%, cement 32%, other additives 7%).

[0139] In this embodiment, by normalizing the reduction percentage and design complexity, and combining fuzzy relation matrix and analytic hierarchy process to perform weighted comprehensive evaluation of multi-dimensional factors, it is possible not only to objectively quantify the performance level of different design schemes, but also to fully consider the balance between performance improvement and design complexity. By comparing performance scores with historical targets, intelligent screening and decision-making on the merits of design schemes are achieved, effectively improving the scientificity and reliability of grounding electrode design scheme evaluation and optimization, and ensuring that the output optimal design scheme achieves the best balance between performance and complexity.

[0140] A second aspect of the present invention provides an electronic device, comprising:

[0141] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0142] A third aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0143] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0144] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A high-efficiency grounding electrode resistance reduction performance optimization design method based on multi-material composite, characterized in that, include: Soil environmental parameters of the grounding electrode installation area are obtained and features are extracted through a deep neural network to generate soil feature vectors. The correlation between different types of data in the soil environmental parameters is calculated by a transformer encoder combined with a multi-head attention mechanism. Based on the correlation, the initial prediction data of resistance reduction performance is determined. Based on the initial prediction data of resistance reduction performance and the soil environmental feature vectors, the grounding electrode optimization parameters are solved by particle swarm optimization. Using the grounding electrode optimization parameters as constraints, a deep deterministic gradient algorithm is employed to calculate the material mix parameters. Minimizing the predicted equivalent resistivity is set as the objective function for dynamic adjustment. Historical data from the experience replay buffer is used to optimize the material mix parameters. Based on the optimized material mix parameters, the actual equivalent resistivity is calculated, including: Obtain the grounding electrode optimization parameters, establish a mass percentage vector of material proportion parameters based on the grounding electrode optimization parameters, construct the material proportion feasible region with the grounding electrode optimization parameters as constraints, and take the sum of the mass percentage vectors as a constant value and the individual material proportions within a preset range as constraints of the material proportion feasible region. A mapping function between material ratio parameters and equivalent resistivity is constructed. The mass percentage vector is mapped to the predicted equivalent resistivity value through the rectified linear unit function based on multilayer perceptron in the mapping function. A state space is constructed based on the mass percentage vector and environmental parameters, and the material ratio adjustment amount is generated as the action space. The reward function value is calculated based on the predicted equivalent resistivity value, the degree of constraint violation, and the difference in material ratio between adjacent time points. The updated mass percentage vector is obtained by adjusting the material ratio parameters according to the reward function value. The state space, action space, reward function value and updated mass percentage vector are stored in the experience replay buffer. Based on the historical data in the experience replay buffer, the optimized material ratio parameters are obtained through the policy gradient method and time difference error. The optimized material ratio parameters are subjected to dynamic noise attenuation and soft parameter update to obtain the optimal material ratio parameters. The actual equivalent resistivity value is calculated by combining the geometric parameters in the grounding electrode optimization parameters. Calculate the percentage reduction of the actual equivalent resistivity value relative to the optimal equivalent resistivity value in the historical optimization record, statistically analyze the range of material ratio parameters in the current design scheme and evaluate the design complexity, use the percentage reduction and design complexity as the evaluation basis to calculate the performance evaluation result, if the performance evaluation result meets the preset performance target, then output the optimized grounding electrode parameters, the actual equivalent resistivity value and the optimized material ratio parameters as the optimal design scheme.

2. The method according to claim 1, characterized in that, Soil environmental parameters of the grounding electrode installation area are obtained, and features are extracted using a deep neural network to generate soil feature vectors. The correlation between different types of data in the soil environmental parameters is then calculated using a transformer encoder combined with a multi-head attention mechanism, including: Soil environmental parameters of the grounding electrode installation area are obtained. Soil resistivity data are collected at different depths using the Wenner quadrupole method. The apparent resistivity data matrix is ​​calculated based on the measured current, voltage values ​​and electrode spacing. Soil dielectric constant is measured using a time-domain reflectometry probe and soil moisture content data is calculated. Geological stratification data is obtained through vertical electrical sounding. The apparent resistivity data matrix, soil moisture content data, and geological stratification data are standardized and recombined into three-channel input data. Convolution and pooling operations are performed on the three-channel input data to obtain soil feature vectors. Location encoding information is generated based on the data acquisition depth and sampling interval and added to the soil feature vectors, which are then divided into multiple feature groups. The query matrix, key matrix, and value matrix of each feature group are calculated using a transformer encoder. The correlation between different types of data in soil environmental parameters is calculated using a multi-head attention mechanism.

3. The method according to claim 1, characterized in that, Based on the correlation, initial prediction data for resistivity reduction performance is determined. Based on this initial prediction data and the soil environmental feature vector, the optimal parameters for the grounding electrode are solved using a particle swarm optimization algorithm, including: Based on the aforementioned correlation, an association matrix is ​​constructed. The weight coefficient of each type of data in the soil environmental parameters is calculated using the standardized mutual information method. An exponential coupling function between different types of data is established in combination with soil physical properties, and the initial prediction data of drag reduction performance is obtained by combining the weight coefficient. An optimization objective function is constructed based on the initial prediction data of drag reduction performance and the soil environment feature vector. The position and velocity of the particle swarm are initialized based on the geometric and material parameters corresponding to the grounding electrode. The optimization objective function is solved by an adaptive mutation particle swarm algorithm. The mutation probability is calculated based on the number of iterations to update the particle position. The fitness value of each particle after the update is calculated and the particles are divided into an exploration group and a development group based on the fitness value. The migration probability between the exploration group and the development group is calculated and the particle distribution is updated based on the migration probability. The particle with the highest fitness value in the updated particle distribution is selected for local search to obtain a local optimum. If the fitness value of the local optimum is greater than the fitness value of the current global optimum, the global optimum position is updated. The mean square error of the particle swarm fitness value and the Lyapunov function value are calculated. When the mean square error is less than the preset mean square error threshold and the Lyapunov function value is monotonically decreasing, the global optimum position is decoded into grounding electrode optimization parameters for output.

4. The method according to claim 1, characterized in that, Based on historical data in the experience replay buffer, optimized material ratio parameters are obtained through the policy gradient method and temporal difference error. The optimized material ratio parameters are then subjected to dynamic noise attenuation and soft parameter updates to obtain the optimal material ratio parameters, including: Historical data is randomly sampled from the experience replay buffer to obtain a transition quadruple consisting of the state space, action space, reward value, and next state space corresponding to the historical data. The reward value at the current moment is calculated based on the transition quadruple and the pre-set reward weight coefficient. A policy function is constructed based on the state space and the action space. The gradient value of the policy function with respect to the material ratio parameter is calculated. Based on the gradient value and the pre-set learning rate, the parameter update amount is calculated in combination with the preset action value function, and the policy function parameters are updated. Based on the current reward value and the action value function, the temporal difference error is calculated in the next state space. The value function parameters are updated according to the temporal difference error and the learning rate. The optimized material ratio parameters are calculated based on the updated policy function parameters and value function parameters. The optimized material ratio parameters are superimposed with exploration noise to obtain noisy material ratio parameters. The noisy material ratio parameters are dynamically attenuated, and the dynamically attenuated material ratio parameters are softly updated using a preset soft update coefficient to obtain the optimal material ratio parameters.

5. The method according to claim 1, characterized in that, Calculate the percentage reduction in actual equivalent resistivity relative to the optimal equivalent resistivity in historical optimization records, and statistically analyze the range of material proportioning parameters in the current design scheme and assess design complexity, including: Obtain the actual equivalent resistivity value in the current design scheme, read the optimal equivalent resistivity value from the historical optimization record, calculate the difference between the optimal equivalent resistivity value and the actual equivalent resistivity value, and calculate the reduction percentage based on the difference. Extract the material ratio parameters from the current design scheme, calculate the Shannon entropy corresponding to the material ratio parameters, calculate the sum of the absolute values ​​of the differences between each pair of different types of parameters in the material ratio parameters, and calculate the Gini coefficient by combining the sum and the parameter type. The material proportioning parameters are calculated, and a parameter correlation matrix is ​​constructed by calculating the covariance and standard deviation. The eigenvalues ​​of the parameter correlation matrix are calculated to obtain the maximum eigenvalue. The design complexity of the current design scheme is calculated based on the Shannon entropy, the Gini coefficient, and the maximum eigenvalue.

6. The method according to claim 1, characterized in that, The performance evaluation results are calculated based on the percentage reduction and design complexity. If the performance evaluation results meet the preset performance targets, the optimized grounding electrode parameters, actual equivalent resistivity values, and optimized material ratio parameters are output as the optimal design scheme, including: The percentage reduction and design complexity are normalized respectively. Based on the normalization results, an evaluation factor set and an evaluation level set are constructed. Based on the membership degree of each factor in the evaluation factor set to each level in the evaluation level set, a fuzzy relation matrix is ​​constructed. The weight values ​​of each factor in the evaluation factor set are calculated using the analytic hierarchy process and a factor weight vector is constructed. The factor weight vector is multiplied by the fuzzy relation matrix to obtain a comprehensive evaluation vector. Each component in the comprehensive evaluation vector is multiplied by the score of the corresponding level and summed to obtain a performance score. The performance target is calculated based on the pre-obtained highest historical performance score and the standard deviation of historical performance scores. Based on the performance score, it is determined whether the current design scheme meets the performance target. If it does, the grounding electrode optimization parameters, the actual equivalent resistivity value, and the optimized material ratio parameters are combined into the optimal design scheme and output.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images