Optimization design method, preparation method, system and equipment of copper-tungsten electrical contact material

Abstract

The invention provides an optimal design method, a preparation method, a system and equipment of a copper-tungsten electrical contact material, which are applied to the technical field of electrical materials. The method comprises the following steps: modeling an optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material based on a deep reinforcement learning algorithm, and constructing an optimization strategy model by taking cumulative reward maximization corresponding to an optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as a target; and training the optimization strategy model based on the obtained typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data so as to optimize each preparation parameter of the copper-tungsten electrical contact material, and obtaining an optimized preparation parameter combination of the copper-tungsten electrical contact material. The problems that comprehensive improvement of multi-dimensional performance of the copper-tungsten composite material is difficult to achieve, the experimental period is long, and manpower and material resources are consumed are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical materials, and in particular to an optimization design method and a preparation method, system and equipment for a copper-tungsten electrical contact material. Background Art

[0002] The localized research, development and application of high-voltage and high-current circuit breaker equipment are of great significance to ensuring the construction of my country's ultra-high voltage AC ring network projects, alleviating the problem of excessive short circuits in ultra-high voltage hybrid power grids, and ensuring the safe and stable operation of large power grids.

[0003] Copper-tungsten composite materials are widely used in power equipment such as high-voltage circuit breakers. However, in order to meet increasingly stringent working conditions, the performance of copper-tungsten composite materials needs to be optimized to improve the performance of equipment and extend the service life of equipment. The existing method of optimizing material performance through experimental exploration often improves the performance of some materials while deteriorating the performance of other materials. It is difficult to improve the comprehensive performance of materials in multiple dimensions, and the experimental cycle is long and consumes manpower and material resources. Summary of the invention

[0004] In order to overcome the problems that the above-mentioned copper-tungsten composite materials are difficult to achieve comprehensive improvement in multi-dimensional performance, have a long experimental cycle, and consume manpower and material resources, the present invention provides an optimization design method and preparation method, system, and equipment for copper-tungsten electrical contact materials.

[0005] In one aspect, the present invention provides a method for optimizing the design of copper-tungsten electrical contact materials, comprising:

[0006] Based on the deep reinforcement learning algorithm, the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material is modeled, and the optimization strategy model is constructed with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material;

[0007] The optimization strategy model is trained based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0008] Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

[0009] Optionally, the microstructure data includes a microstructure image, and the physical property data includes electrical conductivity, hardness, wear resistance and arc erosion resistance.

[0010] Optionally, the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material is modeled based on the deep reinforcement learning algorithm, and the optimization strategy model is constructed with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material, including:

[0011] Taking the preparation parameters of the copper-tungsten electrical contact material as the observed state values ​​of the intelligent agent;

[0012] Using the adjustment amount of each preparation parameter of the copper-tungsten electrical contact material as the action of the intelligent agent;

[0013] Determining a reward for an action based on multi-dimensional performance data corresponding to various preparation parameters of the copper-tungsten electrical contact material;

[0014] Taking the maximization of the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the objective function, the optimization strategy model is constructed based on a deep neural network to determine the value of the action.

[0015] Optionally, the optimization strategy model is trained based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material to obtain the optimized preparation parameter combination of the copper-tungsten electrical contact material, including:

[0016] Based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the interaction between the intelligent agent and the environment is simulated by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to obtain the state transition data corresponding to each action;

[0017] Store the state transition data corresponding to each action into the experience buffer;

[0018] Based on the data sampling of the experience buffer, a corresponding training sample is obtained;

[0019] Training the optimization strategy model based on the training samples;

[0020] The state after the action at the end of the training is performed is used as the optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0021] In which, during the training process, the model parameters of the optimization strategy model are updated based on the mean square error between the value of the current action and the target value; if the training is completed, the target value is the reward corresponding to the current action, otherwise, the target value is determined based on the reward corresponding to the current action and the discounted reward of the next state after the current action is executed.

[0022] Optionally, based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the multi-dimensional performance data of the copper-tungsten electrical contact material is simulated to perform the interaction between the intelligent agent and the environment, and the state transition data corresponding to each action is obtained, including:

[0023] Select an action as the current action based on the current state;

[0024] By simulating the multi-dimensional performance data of copper-tungsten electrical contact materials, the intelligent agent interacts with the environment to perform the current action, and determines the performance data of each dimension and the next state after the current action is executed;

[0025] Based on the indicator weights corresponding to the performance data of each dimension, the deviation amount of the performance data of each dimension is weighted and fused to determine the reward corresponding to the current action;

[0026] The current state, current action, reward corresponding to the current action, and next state are used as the state transition data corresponding to the current action;

[0027] Among them, the indicator weights corresponding to the performance data of each dimension belong to the model parameters of the optimization strategy model, and the deviation amount of the performance data of each dimension is the deviation between the performance data of each dimension and the expected value of the performance of each dimension after the current action is executed.

[0028] Optionally, the performance data of each dimension includes a microstructure image, and the interaction between the intelligent agent and the environment is performed by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to perform the current action, and the performance data of each dimension after the current action is performed is determined, including:

[0029] The microstructure of the copper-tungsten electrical contact material is generated by simulation to interact with the environment to perform the current action, and the microstructure image after the current action is performed is obtained;

[0030] Based on the pre-trained feature extraction network, feature extraction is performed on the microstructure image after the current action is executed to obtain the microscopic feature vector after the current action is executed.

[0031] Optionally, before performing weighted fusion processing on the deviation amount of the performance data of each dimension based on the indicator weight corresponding to the performance data of each dimension to determine the reward corresponding to the current action, the method further includes:

[0032] Based on the pre-trained feature extraction network, feature extraction is performed on the preset standard microstructure image to obtain the target feature vector;

[0033] Based on the distance between the micro-feature vector after the current action is executed and the target feature vector, the current micro-performance deviation is determined.

[0034] Optionally, the features of the microstructure image include the distribution of the reinforcement phase of the copper-tungsten electrical contact material, the microscopic morphology of the copper-tungsten electrical contact material, and the grain boundary structure of each crystal phase in the copper-tungsten electrical contact material.

[0035] On the other hand, the present invention also provides a method for preparing a copper-tungsten electrical contact material, comprising:

[0036] The copper-tungsten electrical contact material is prepared based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained based on any one of the above items.

[0037] On the other hand, the present invention also provides an optimization design system for copper-tungsten electrical contact materials, comprising:

[0038] A model building module, for modeling the optimization adjustment process of the preparation parameters of the copper-tungsten electrical contact material based on a deep reinforcement learning algorithm, and building an optimization strategy model with the goal of maximizing the cumulative reward corresponding to the optimization strategy of the preparation parameters of the copper-tungsten electrical contact material;

[0039] An optimization module, used for training the optimization strategy model based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, so as to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0040] Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

[0041] On the other hand, the present invention also provides a system for preparing a copper-tungsten electrical contact material, comprising:

[0042] A preparation module is used to prepare the copper-tungsten electrical contact material based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained in any one of the above items.

[0043] On the other hand, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0044] The memory is used to store one or more programs;

[0045] When the one or more programs are executed by the at least one processor, any one of the above methods is implemented.

[0046] On the other hand, the present invention further provides a readable storage medium having an execution program stored thereon, and when the execution program is executed, any one of the methods described above is implemented.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention provides an optimization design method for copper-tungsten electrical contact materials. Based on a deep reinforcement learning algorithm, an optimization adjustment process of various preparation parameters of the copper-tungsten electrical contact materials is modeled. Taking maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the goal, an optimization strategy model is constructed and the optimization strategy model is trained. Automatic optimization adjustment of the preparation parameters of the copper-tungsten electrical contact materials is achieved through deep reinforcement learning. No experimental exploration is required, the optimization efficiency is high, and manpower and material resources are saved.

[0049] The present invention determines the optimization strategy reward through the multi-dimensional performance data of the copper-tungsten electrical contact material, and realizes the optimization adjustment of multi-dimensional preparation parameters such as the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters guided by multi-dimensional performance data such as microstructure data and physical properties. The combination of optimized preparation parameters based on the copper-tungsten electrical contact material can achieve the synergistic improvement of the multi-dimensional performance of the copper-tungsten electrical contact material. At the same time, by considering the multi-dimensional preparation parameters, the performance bottleneck problem caused by the single-dimensional preparation parameters can be avoided; by considering the microstructure data, the optimization guidance of the microstructure is increased, and the stability of the optimized performance is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of a process flow of an optimization design method for a copper-tungsten electrical contact material of the present invention;

[0051] Figure 2 It is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0052] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0053] Embodiment 1:

[0054] The present invention provides an optimization design method for copper-tungsten electrical contact materials, as shown in the schematic diagram Figure 1 As shown, including:

[0055] Step S110, modeling the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material based on a deep reinforcement learning algorithm, and constructing an optimization strategy model with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material;

[0056] Step S120, training the optimization strategy model based on the acquired typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0057] Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

[0058] In this example implementation, the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material is transformed into an interactive learning process between an intelligent agent and an environment through a deep reinforcement learning algorithm, maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material, and completing the optimization adjustment process of the preparation parameters by training the optimization strategy model. Exemplarily, the microstructure data includes a microstructure image, and the physical property data includes electrical conductivity, hardness, wear resistance, and arc erosion resistance. The types of enhanced phases include carbon nanophases, oxides, MAX phases, etc., and the copper-tungsten matrix parameters include the copper-tungsten ratio in the raw material, the particle size, oxygen content, purity, etc. of the raw copper powder and the raw tungsten powder; for example, the typical preparation parameter range of the raw material can be set as: the average particle size of tungsten powder is 4-8μm, the oxygen content is ≤600ppm, and the purity is ≥99.9%, and the particle size of electrolytic copper powder is 30-60μm, the oxygen content is ≤500ppm, and the purity is ≥99.9%. The preparation process parameters include the preparation method and the process parameters corresponding to the preparation method. The preparation method may include melt infiltration method, copper oxide powder method, injection molding method, nano-carbon material modification method, etc. For example, the process parameters of the melt infiltration method include raw material mixing process parameters, embryo molding parameters and melt infiltration process parameters. The typical process parameter range can be set as follows: powder mixing speed 700-1000r / min, powder mixing time: 1-2h, embryo pressure and time: 600-700MPa, 1-2h, melt infiltration temperature 1300-1400℃, time 2-4h. Microstructure data include microstructure images, such as scanning electron microscope images; physical property data include conductivity (such as conductivity), hardness, wear resistance (such as characterized by friction coefficient) and arc ablation resistance (such as characterized by ablation mass loss). The present invention uses a deep Q learning algorithm to design an intelligent agent model for optimizing the preparation parameters of copper-tungsten electrical contact materials. Through interactive learning between the intelligent agent and the environment, the environment can be a performance simulation system of copper-tungsten electrical contact materials constructed based on existing experimental data, which can include microscopic phase field simulation and finite element simulation of physical properties, or a neural network model trained based on existing experimental data. The neural network model aims to establish a mapping relationship between the preparation parameters and performance indicators of copper-tungsten electrical contact materials. Based on this, the structure-activity relationship between the composition design of copper-tungsten electrical contact materials and microstructure and physical properties is mastered; the correlation between the preparation parameters of copper-tungsten electrical contact materials and microscopic properties, arc burning and ablation behavior and electrical characteristics is established. On this basis, the optimal material parameter combination is gradually approached to maximize the performance improvement. The prepared electrical contact material has excellent electrical conductivity and mechanical properties, and is efficient, cheap and environmentally friendly.

[0059] The present invention can intelligently optimize the material parameters and key preparation processes (such as sintering, infiltration, heat treatment, etc.) of copper-tungsten composite materials through a deep Q learning algorithm, and realize the simultaneous optimization of material design and process adjustment. This technology not only optimizes the material composition, but also dynamically adjusts the process parameters during the preparation process to ensure that the comprehensive performance of the final material is optimal. Expanding from a single material parameter to the preparation process greatly improves the optimization efficiency and the actual application performance of the material.

[0060] In an example implementation, the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material based on the deep reinforcement learning algorithm in step S110 is modeled, and the optimization strategy model is constructed with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material, including:

[0061] Taking the preparation parameters of the copper-tungsten electrical contact material as the observed state values ​​of the intelligent agent;

[0062] Using the adjustment amount of each preparation parameter of the copper-tungsten electrical contact material as the action of the intelligent agent;

[0063] Determining a reward for an action based on multi-dimensional performance data corresponding to various preparation parameters of the copper-tungsten electrical contact material;

[0064] Taking the maximization of the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the objective function, the optimization strategy model is constructed based on a deep neural network to determine the value of the action.

[0065] In this example embodiment, the reward of the action is determined by interaction with the environment, and each action corresponds to an adjustment to a material preparation parameter (e.g., increase, decrease, or remain unchanged), and the reward reflects the effect of the parameter adjustment on the material performance. Each selected action can be understood as an adjustment to a parameter. For example, increasing a parameter value (e.g., increasing the copper content, increasing the sintering temperature, etc.). The state change after the action selection will affect the performance of the material, and the environment determines the reward value by simulating the performance change brought about by the action. The optimization strategy model is a DQN (Deep Q-Network) model, which may include an input layer, a hidden layer, and an output layer, wherein the number of neurons in the input layer is equal to the dimension of the state space, that is, the number of preparation parameters of the copper-tungsten electrical contact material (e.g., the sum of the parameters of the type of enhanced phase, copper-tungsten matrix parameters, and process parameters), and the input layer is used to receive the current state and pass it to the subsequent hidden layer. The hidden layer can be a network structure with no less than two layers, each layer having no less than 24 neurons, and using a ReLU (Rectified Linear Unit) activation function, which helps the model capture nonlinear relationships, so that the neural network can better learn the complex relationship between material preparation parameters and their performance. The number of neurons in the output layer is equal to the dimension of the action space, that is, the possible adjustment amount of each material parameter. A positive adjustment amount indicates an increase, a negative adjustment amount indicates a decrease, and an adjustment amount of 0 indicates that the parameter remains unchanged. The output layer uses a linear activation function to output the value of each action (i.e., Q value) to evaluate the pros and cons of different actions. In other words, the input of the optimization strategy model corresponds to the observed state value of the agent, and the output corresponds to the value of the action.

[0066] In an exemplary embodiment, the step S120 of training the optimization strategy model based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain the optimized preparation parameter combination of the copper-tungsten electrical contact material includes:

[0067] Based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the interaction between the intelligent agent and the environment is simulated by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to obtain the state transition data corresponding to each action;

[0068] Store the state transition data corresponding to each action into the experience buffer;

[0069] Based on the data sampling of the experience buffer, a corresponding training sample is obtained;

[0070] Training the optimization strategy model based on the training samples;

[0071] The state after the action at the end of the training is performed is used as the optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0072] In which, during the training process, the model parameters of the optimization strategy model are updated based on the mean square error between the value of the current action and the target value; if the training is completed, the target value is the reward corresponding to the current action, otherwise, the target value is determined based on the reward corresponding to the current action and the discounted reward of the next state after the current action is executed.

[0073] In this example embodiment, an environment can be built based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data. The environment can be a simulation system for simulating the multi-dimensional performance data of the copper-tungsten electrical contact material, or a neural network model that establishes a relationship between the preparation parameters and the performance indicators. For example, the hardness, wear resistance and other tests of the material are simulated by finite element simulation to obtain the simulated performance indicators. The typical preparation parameters of the copper-tungsten electrical contact material can also be used as the initial state parameters of the environment. In order to train the optimization strategy model, i.e., the deep neural network, the agent can randomly extract small batches of data from the experience buffer as training samples for training. Random sampling helps to break the correlation between experiences and improve the generalization ability of the model. Through experience replay, the agent can reuse the samples in the experience buffer at each time step to enhance the learning effect. During each training, the agent obtains action rewards through interaction with the environment, and the agent calculates the target Q value and updates the predicted value of the neural network. Specifically, if the current action leads to the end of training, the immediate reward of the current action is used as the target Q value; otherwise, the sum of the immediate reward of the current action and the discounted maximum Q value of the next state is used as the target Q value. The model uses mean-square error (MSE) as the loss function, that is, the loss function is defined by the difference between the Q value and the target Q value. Since the goal of deep Q learning is to let the agent learn to choose actions that can maximize future cumulative rewards in a certain state (i.e., a set of material parameter combinations). Therefore, the loss function is used to measure the difference between the current Q value estimate and the ideal target Q value. For example, MSE is used to calculate the error between the predicted current Q value and the target Q value, and the formula is as follows:

[0074]

[0075] Where L represents the loss function, N is the number of samples in the experience buffer, i is the sample index of the experience buffer, and Q i is the predicted Q value of the i-th sample, is the target Q value of the i-th sample. Use the Adam optimizer to update the model parameters. Repeat the above model training process until the model converges and the training is ended. The state after the action is executed at the end of the training is used as the optimized preparation parameter combination of the copper-tungsten electrical contact material. MSE is selected in this example because it is simple and effective, and is suitable for regression problems. The Adam optimizer can dynamically adjust the learning rate to ensure that the model converges smoothly during the training process. During the training process of the model, the intelligent agent continuously accumulates experience through interaction with the environment and optimizes the material preparation parameter combination.

[0076] Exemplarily, based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the multi-dimensional performance data of the copper-tungsten electrical contact material is simulated to perform the interaction between the intelligent agent and the environment, and the state transition data corresponding to each action is obtained, including:

[0077] Select an action as the current action based on the current state;

[0078] By simulating the multi-dimensional performance data of copper-tungsten electrical contact materials, the intelligent agent interacts with the environment to perform the current action, and determines the performance data of each dimension and the next state after the current action is executed;

[0079] Based on the indicator weights corresponding to the performance data of each dimension, a weighted fusion process is performed on the change amount of the performance data of each dimension corresponding to the current action to determine the reward corresponding to the current action;

[0080] The current state, current action, reward corresponding to the current action, and next state are used as the state transition data corresponding to the current action;

[0081] Among them, the indicator weights corresponding to the performance data of each dimension belong to the model parameters of the optimization strategy model, and the deviation amount of the performance data of each dimension is the deviation between the performance data of each dimension and the expected value of the performance of each dimension after the current action is executed.

[0082] In this example implementation, at the beginning of each round of training, the initial state is selected, and the agent selects the action using the ε-greedy strategy, that is, the agent randomly selects the action under a certain probability (i.e., the exploration phase), and the environment simulates the corresponding performance data according to the action, determines the action reward and the next state by the deviation between the current performance and the expected performance, and completes the state transfer. The expected value of each dimension performance can be determined based on the application scenario of the copper-tungsten electrical contact material. For example, the application scenario is the SF6 circuit breaker for 110kV and above high-voltage and ultra-high-voltage transmission and transformation systems. The ideal high-voltage electrical contact material requires large current breaking capacity, withstand voltage, small contact resistance, good resistance to welding, wear resistance, small cutoff current, high mechanical strength and good processing performance. For example, if the microstructure expectation is that the uniformity of the enhanced phase distribution is good, the corresponding uniformity index can be designed to characterize the distribution of the enhanced phase. In order to achieve a balance between exploration and utilization of the agent, the exploration rate ε can be gradually decayed from 1 to 0.01. After the ε value is reduced to a certain value (such as 0.01), the action with the highest Q value is selected in the model training process (i.e., the utilization phase). As the training progresses, ε gradually decays, and the agent will choose more optimal actions, reduce randomness, and thus accelerate convergence to improve performance. The state transition data corresponding to the current action includes the current state, the current action, the reward of the current action, and the next state. Whether the current training is finished can also be stored in the experience buffer. For example, after each action is executed, the agent will record the current state, the action performed, the reward obtained, the next state, and whether the training is finished, and store this information in the experience buffer for subsequent training. Exemplarily, the environment performs finite element analysis on the macroscopic physical properties of the material based on the preparation parameters provided by the agent to obtain the current multi-dimensional physical properties, which include conductivity, hardness, wear resistance, and arc erosion resistance. This example can evaluate and optimize the comprehensive properties of the material such as conductivity, hardness, wear resistance, and arc erosion resistance from a macroscopic scale, ensure the macroscopic performance of the material, avoid inrush current erosion several times higher than the rated current during multiple operations of the SF6 circuit breaker, and avoid contact deformation caused by mechanical wear between the moving and static arc contacts, and solve the problem that the metal vapor generated by the contact deformation is easy to damage the insulation performance of the arc extinguishing chamber.

[0083] Exemplarily, for each dimensional performance data including a microstructure image, the interaction between the intelligent agent and the environment is performed by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to perform the current action, and the performance data of each dimension after the current action is performed is determined, including:

[0084] The microstructure of the copper-tungsten electrical contact material is generated by simulation to interact with the environment to perform the current action, and the microstructure image after the current action is performed is obtained;

[0085] Based on the pre-trained feature extraction network, feature extraction is performed on the microstructure image after the current action is executed to obtain the microscopic feature vector after the current action is executed.

[0086] In this example implementation, the environment can perform phase field simulation on the microstructure of the material based on the preparation parameters provided by the intelligent agent. For the microstructure of the material, the microscopic simulation is achieved by simulating the key information such as the distribution of the reinforcement phase, micromorphology, and grain boundary structure; different microstructural indicators can also be set, and then the pre-trained neural network model can be used to predict each microstructural indicator. The microstructural indicators may include the reinforcement phase distribution uniformity index, the average grain size, and the grain boundary length index. Microscopic properties affect the material's conductivity, hardness, wear resistance, and arc erosion resistance. During the optimization process, microscopic-scale simulations are used to provide key information about the impact of the material's internal structure on its performance. By combining microscopic and macroscopic simulations, the system can comprehensively evaluate the material to ensure that the optimized parameters have good performance in practical applications. If the microstructure is a microstructure image, in order to enable the microstructure image to be calculated and processed together with other index-type physical performance data, the microstructure image after the action is executed can be obtained through the interaction between the intelligent agent and the environment, and the microstructure image is extracted by a pre-trained feature extraction network (such as a convolutional neural network) to obtain the corresponding feature vector, and the performance change and microstructure change of the microstructure dimension are determined by calculating the distance between the corresponding feature vectors of the microstructure image before and after the action is executed (such as Euclidean distance, etc.). The feature extraction network can be pre-trained with a large number of scanning electron microscope images of copper-tungsten electrical contact materials so that the network can accurately extract features. For example, the key information of the microstructure can be obtained by extracting the distribution of the reinforcing phase, the micromorphological characteristics and the grain boundary structure of each crystal phase in the microstructure image of the copper-tungsten electrical contact material.

[0087] Exemplarily, in order to calculate the deviation of the microstructure, before weighted fusion processing is performed on the deviation of the performance data of each dimension based on the indicator weight corresponding to the performance data of each dimension and determining the reward corresponding to the current action, the method further includes:

[0088] Based on the pre-trained feature extraction network, feature extraction is performed on the preset standard microstructure image to obtain the target feature vector;

[0089] Based on the distance between the micro-feature vector after the current action is executed and the target feature vector, the current micro-performance deviation is determined.

[0090] In this example implementation, the standard microstructure image is a microstructure image of an ideal copper-tungsten electrical contact material. The standard microstructure image is input into a pre-trained feature extraction network for feature extraction to obtain a corresponding target feature vector. The current microscopic performance deviation is determined by calculating the distance (such as the Euclidean distance) between the microscopic feature vector after the current action is executed and the target feature vector.

[0091] For example, the performance data of each dimension includes microstructure and multiple physical properties. Let the microscopic properties of the material be m and each physical property be P. j After the agent performs the action, the deviation between the current microscopic performance and the standard microscopic image is Δm, and the deviation between the physical performance of each dimension and the expected performance is ΔP j , then the reward of the current action can be expressed as:

[0092]

[0093] Among them, R is the reward of the current action, w m is the index weight of microstructure, w j is the indicator weight of the jth physical performance dimension, n is the number of physical performance dimensions and j is the physical performance dimension index. The estimated Q value is updated using the reward of the current action.

[0094] The multi-dimensional performance data of this example includes microstructure and multiple physical properties. It mainly considers multiple scales to comprehensively evaluate the material at the micro and macro scales. Through phase field simulation of microstructure and finite element analysis of macro mechanical properties, the microstructure focuses on the distribution of the reinforcing phase, the size effect and the influence of morphology on material properties, while the macro scale focuses on evaluating the performance of the material in key performance indicators such as conductivity, hardness and wear resistance. This can form an intelligent optimization closed loop from microstructure to macro performance, ensuring the consistency and accuracy of the optimization scheme at multiple scales. It realizes the precise linkage optimization of microstructure and macro performance, makes up for the shortcomings of traditional single-scale optimization methods, and provides a more comprehensive means of material performance evaluation.

[0095] Based on deep Q-learning (DQN), the present invention proposes a new method for automatically adjusting material parameters through intelligent algorithms. Traditional optimization methods often rely on experiments and experience, while this solution uses reinforcement learning technology to realize the intelligence and automation of material design, greatly improving efficiency. The present invention combines multi-scale simulation with deep Q-learning, breaking through the limitations of previous methods in multi-objective and multi-dimensional performance optimization. Multi-scale simulation provides a comprehensive evaluation of the impact of microstructure on macroscopic performance, while deep Q-learning generates optimal strategies through continuous interactive learning. This combination makes material optimization more systematic and scientific. The present invention allows deep Q-learning agents to balance between multiple performance indicators. Traditional methods can only manually adjust and weigh the relationship between performance, avoiding the limitations of single indicator optimization in traditional optimization methods. This solution can automatically explore the optimal solution, greatly improving the efficiency of multi-objective optimization. The experience replay mechanism ensures that the model can repeatedly use existing data during training, improves the learning ability of the agent, and enables it to converge quickly in complex parameter spaces. The agent model using deep Q learning can automatically perform intelligent search and optimization in large-scale parameter spaces, which has significant innovative significance. While conducting research on copper-tungsten materials, the present invention provides theoretical support for the future field of material design.

[0096] The present invention can realize the collaborative optimization of copper-tungsten composite materials under multiple objectives through the combination of deep Q learning intelligent agents and multi-scale simulation. The first is dynamic material parameter adjustment, that is, the intelligent agent dynamically explores the optimal combination of material performance by adjusting the type of reinforcing phase, copper-tungsten matrix parameters and key process parameters. Secondly, the comprehensive improvement of multiple performances, that is, the optimization is not only for a single performance indicator, but also comprehensively considers conductivity, hardness, wear resistance and arc erosion resistance, so that the optimized material can be used under a variety of working conditions. Another is the adaptive optimization path, that is, the intelligent agent adaptively adjusts the optimization path in each round of optimization process, and gradually approaches the optimal material combination. The method of the present invention constructs a reinforcement learning intelligent agent to dynamically adjust and optimize in a high-dimensional material parameter space, covering multi-dimensional optimization strategies such as reinforcing phase types, material composition, and process parameters to improve the key properties of the material such as conductivity, hardness, wear resistance and arc erosion resistance. The collaborative improvement of multi-dimensional performance and the intelligent search for the optimal material parameter combination are realized.

[0097] Example 2

[0098] Based on the same inventive concept, the present invention also provides a method for preparing a copper-tungsten electrical contact material, comprising:

[0099] The copper-tungsten electrical contact material is prepared based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained in any implementation manner in Example 1.

[0100] The copper-tungsten electrical contact material prepared in this example has excellent conductivity, high hardness, good wear resistance and outstanding arc erosion resistance, and can be widely used in power equipment such as high-voltage circuit breakers. The optimized material not only has higher performance, but can also be customized according to specific application requirements to improve the stability and service life of the equipment.

[0101] Example 3

[0102] Based on the same inventive concept, the present invention also provides an optimization design system for copper-tungsten electrical contact materials, comprising:

[0103] A model building module, for modeling the optimization adjustment process of the preparation parameters of the copper-tungsten electrical contact material based on a deep reinforcement learning algorithm, and building an optimization strategy model with the goal of maximizing the cumulative reward corresponding to the optimization strategy of the preparation parameters of the copper-tungsten electrical contact material;

[0104] An optimization module, used for training the optimization strategy model based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, so as to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material;

[0105] Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

[0106] In a possible implementation, the microstructure data includes a microstructure image, and the physical property data includes electrical conductivity, hardness, wear resistance, and arc erosion resistance.

[0107] In a possible implementation, the model building module is specifically used to:

[0108] Taking the preparation parameters of the copper-tungsten electrical contact material as the observed state values ​​of the intelligent agent;

[0109] Using the adjustment amount of each preparation parameter of the copper-tungsten electrical contact material as the action of the intelligent agent;

[0110] Determining a reward for an action based on multi-dimensional performance data corresponding to various preparation parameters of the copper-tungsten electrical contact material;

[0111] Taking the maximization of the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the objective function, the optimization strategy model is constructed based on a deep neural network to determine the value of the action.

[0112] In a possible implementation, the optimization module includes:

[0113] An interaction submodule, for performing interaction between an intelligent agent and an environment by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material based on typical preparation parameters of the copper-tungsten electrical contact material and corresponding multi-dimensional performance data, and obtaining state transition data corresponding to each action;

[0114] The cache submodule is used to store the state transition data corresponding to each action into the experience buffer;

[0115] A sampling submodule, configured to obtain corresponding training samples based on sampling of data in the experience buffer;

[0116] A training submodule, for training the optimization strategy model based on the training samples; using the state after the action is executed at the end of the training as the optimization preparation parameter combination of the copper-tungsten electrical contact material;

[0117] In which, during the training process, the model parameters of the optimization strategy model are updated based on the mean square error between the value of the current action and the target value; if the training is completed, the target value is the reward corresponding to the current action, otherwise, the target value is determined based on the reward corresponding to the current action and the discounted reward of the next state after the current action is executed.

[0118] In a possible implementation, the interaction submodule includes:

[0119] an action selection subunit, for selecting an action as a current action based on a current state;

[0120] An action execution subunit is used to perform the interaction between the intelligent agent and the environment by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to execute the current action, and determine the performance data of each dimension and the next state after the current action is executed;

[0121] A reward determination subunit, used to perform weighted fusion processing on the change amount of each dimension performance data corresponding to the current action based on the indicator weight corresponding to each dimension performance data, and determine the reward corresponding to the current action;

[0122] A state transfer determination subunit, used to use the current state, the current action, the reward corresponding to the current action, and the next state as the state transfer data corresponding to the current action;

[0123] Among them, the indicator weights corresponding to the performance data of each dimension belong to the model parameters of the optimization strategy model, and the deviation amount of the performance data of each dimension is the deviation between the performance data of each dimension and the expected value of the performance of each dimension after the current action is executed.

[0124] In a possible implementation manner, the performance data of each dimension includes a microstructure image, and the action execution subunit is specifically used for:

[0125] The microstructure of the copper-tungsten electrical contact material is generated by simulation to interact with the environment to perform the current action, and the microstructure image after the current action is performed is obtained;

[0126] Based on the pre-trained feature extraction network, feature extraction is performed on the microstructure image after the current action is executed to obtain the microscopic feature vector after the current action is executed.

[0127] In a possible implementation manner, the reward determination subunit is further configured to:

[0128] Based on the pre-trained feature extraction network, feature extraction is performed on the preset standard microstructure image to obtain the target feature vector;

[0129] Based on the distance between the micro-feature vector after the current action is executed and the target feature vector, the current micro-performance deviation is determined.

[0130] In a possible implementation, the features of the microstructure image include the distribution of the reinforcement phase of the copper-tungsten electrical contact material, the microscopic morphology of the copper-tungsten electrical contact material, and the grain boundary structure of each crystal phase in the copper-tungsten electrical contact material.

[0131] Example 4

[0132] Based on the same inventive concept, the present invention also provides a system for preparing copper-tungsten electrical contact materials, comprising:

[0133] A preparation module is used to prepare the copper-tungsten electrical contact material based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained in any implementation manner of Example 1.

[0134] Example 5

[0135] like Figure 2 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0136] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of an optimization design method and preparation method of a copper-tungsten electrical contact material in the above-mentioned embodiment.

[0137] Example 6

[0138] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in electronic devices and, of course, extended storage media supported by electronic devices. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of an optimization design method and preparation method of a copper-tungsten electrical contact material in the above embodiment.

[0139] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for optimizing the design of copper-tungsten electrical contact materials, characterized in that: include: Based on the deep reinforcement learning algorithm, the optimization adjustment process of each preparation parameter of the copper-tungsten electrical contact material is modeled, and the optimization strategy model is constructed with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material; The optimization strategy model is trained based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material; Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

2. The method according to claim 1, characterized in that The microstructure data includes a microstructure image, and the physical property data includes electrical conductivity, hardness, wear resistance and arc erosion resistance.

3. The method according to claim 1 or 2, characterized in that: The optimization adjustment process of the preparation parameters of the copper-tungsten electrical contact material is modeled based on the deep reinforcement learning algorithm, and the optimization strategy model is constructed with the goal of maximizing the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material, including: Taking the preparation parameters of the copper-tungsten electrical contact material as the observed state values ​​of the intelligent agent; Using the adjustment amount of each preparation parameter of the copper-tungsten electrical contact material as the action of the intelligent agent; Determining a reward for an action based on multi-dimensional performance data corresponding to various preparation parameters of the copper-tungsten electrical contact material; Taking the maximization of the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the objective function, the optimization strategy model is constructed based on a deep neural network to determine the value of the action.

4. The method according to claim 3, characterized in that The optimization strategy model is trained based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain the optimized preparation parameter combination of the copper-tungsten electrical contact material, including: Based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the interaction between the intelligent agent and the environment is simulated by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to obtain the state transition data corresponding to each action; Store the state transition data corresponding to each action into the experience buffer; Based on the data sampling of the experience buffer, a corresponding training sample is obtained; Training the optimization strategy model based on the training samples; The state after the action at the end of the training is performed is used as the optimized preparation parameter combination of the copper-tungsten electrical contact material; In which, during the training process, the model parameters of the optimization strategy model are updated based on the mean square error between the value of the current action and the target value; if the training is completed, the target value is the reward corresponding to the current action, otherwise, the target value is determined based on the reward corresponding to the current action and the discounted reward of the next state after the current action is executed.

5. The method according to claim 4, characterized in that Based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, the multi-dimensional performance data of the copper-tungsten electrical contact material is simulated to perform the interaction between the intelligent agent and the environment, and the state transition data corresponding to each action is obtained, including: Select an action as the current action based on the current state; By simulating the multi-dimensional performance data of copper-tungsten electrical contact materials, the intelligent agent interacts with the environment to perform the current action, and determines the performance data of each dimension and the next state after the current action is executed; Based on the indicator weights corresponding to the performance data of each dimension, the deviation of the performance data of each dimension is weighted and fused to determine the reward corresponding to the current action; The current state, current action, reward corresponding to the current action, and next state are used as the state transition data corresponding to the current action; Among them, the indicator weights corresponding to the performance data of each dimension belong to the model parameters of the optimization strategy model, and the deviation amount of the performance data of each dimension is the deviation between the performance data of each dimension and the expected value of the performance of each dimension after the current action is executed.

6. The method according to claim 5, characterized in that The performance data of each dimension includes a microstructure image, and the interaction between the intelligent agent and the environment is performed by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to perform the current action, and the performance data of each dimension after the current action is performed is determined, including: The microstructure of the copper-tungsten electrical contact material is generated by simulation to interact with the environment to perform the current action, and the microstructure image after the current action is performed is obtained; Based on the pre-trained feature extraction network, feature extraction is performed on the microstructure image after the current action is executed to obtain the microscopic feature vector after the current action is executed.

7. The method according to claim 6, characterized in that Before performing weighted fusion processing on the deviation of each dimension performance data based on the indicator weight corresponding to each dimension performance data and determining the reward corresponding to the current action, the method further includes: Based on the pre-trained feature extraction network, feature extraction is performed on the preset standard microstructure image to obtain the target feature vector; Based on the distance between the micro-feature vector after the current action is executed and the target feature vector, the current micro-performance deviation is determined.

8. The method according to claim 7, characterized in that The features of the microstructure image include the distribution of the reinforcement phase of the copper-tungsten electrical contact material, the microscopic morphology of the copper-tungsten electrical contact material, and the grain boundary structure of each crystal phase in the copper-tungsten electrical contact material.

9. A method for preparing a copper-tungsten electrical contact material, characterized in that: include: The copper-tungsten electrical contact material is prepared based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained according to any one of claims 1 to 8.

10. An optimization design system for copper-tungsten electrical contact materials, characterized in that: include: A model building module, for modeling the optimization adjustment process of the preparation parameters of the copper-tungsten electrical contact material based on a deep reinforcement learning algorithm, and building an optimization strategy model with the goal of maximizing the cumulative reward corresponding to the optimization strategy of the preparation parameters of the copper-tungsten electrical contact material; An optimization module, used for training the optimization strategy model based on the typical preparation parameters of the copper-tungsten electrical contact material and the corresponding multi-dimensional performance data, so as to optimize the preparation parameters of the copper-tungsten electrical contact material and obtain an optimized preparation parameter combination of the copper-tungsten electrical contact material; Among them, the typical preparation parameters include the type of reinforcement phase, copper-tungsten matrix parameters and preparation process parameters, and the multi-dimensional performance data include microstructure data and physical property data; the reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material is determined based on the multi-dimensional performance data of the copper-tungsten electrical contact material.

11. The system according to claim 10, characterized in that The microstructure data includes a microstructure image, and the physical property data includes electrical conductivity, hardness, wear resistance and arc erosion resistance.

12. The system according to claim 10 or 11, characterized in that The model building module is specifically used for: Taking the preparation parameters of the copper-tungsten electrical contact material as the observed state values ​​of the intelligent agent; Using the adjustment amount of each preparation parameter of the copper-tungsten electrical contact material as the action of the intelligent agent; Determining a reward for an action based on multi-dimensional performance data corresponding to various preparation parameters of the copper-tungsten electrical contact material; Taking the maximization of the cumulative reward corresponding to the optimization strategy of each preparation parameter of the copper-tungsten electrical contact material as the objective function, the optimization strategy model is constructed based on a deep neural network to determine the value of the action.

13. The system according to claim 12, characterized in that The optimization module includes: An interaction submodule, for performing interaction between an intelligent agent and an environment by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material based on typical preparation parameters of the copper-tungsten electrical contact material and corresponding multi-dimensional performance data, and obtaining state transition data corresponding to each action; The cache submodule is used to store the state transition data corresponding to each action into the experience buffer; A sampling submodule, configured to obtain corresponding training samples based on sampling of data in the experience buffer; A training submodule, for training the optimization strategy model based on the training samples; using the state after the action is executed at the end of the training as the optimization preparation parameter combination of the copper-tungsten electrical contact material; In which, during the training process, the model parameters of the optimization strategy model are updated based on the mean square error between the value of the current action and the target value; if the training is completed, the target value is the reward corresponding to the current action, otherwise, the target value is determined based on the reward corresponding to the current action and the discounted reward of the next state after the current action is executed.

14. The system according to claim 13, characterized in that The interaction submodule includes: an action selection subunit, for selecting an action as a current action based on a current state; An action execution subunit is used to perform the interaction between the intelligent agent and the environment by simulating the multi-dimensional performance data of the copper-tungsten electrical contact material to execute the current action, and determine the performance data of each dimension and the next state after the current action is executed; A reward determination subunit, used to perform weighted fusion processing on the change amount of each dimension performance data corresponding to the current action based on the indicator weight corresponding to each dimension performance data, and determine the reward corresponding to the current action; A state transfer determination subunit, used to use the current state, the current action, the reward corresponding to the current action, and the next state as the state transfer data corresponding to the current action; Among them, the indicator weights corresponding to the performance data of each dimension belong to the model parameters of the optimization strategy model, and the deviation amount of the performance data of each dimension is the deviation between the performance data of each dimension and the expected value of the performance of each dimension after the current action is executed.

15. The system according to claim 14, characterized in that The performance data of each dimension includes a microstructure image, and the action execution subunit is specifically used for: The microstructure of the copper-tungsten electrical contact material is generated by simulation to interact with the environment to perform the current action, and the microstructure image after the current action is performed is obtained; Based on the pre-trained feature extraction network, feature extraction is performed on the microstructure image after the current action is executed to obtain the microscopic feature vector after the current action is executed.

16. The system according to claim 15, characterized in that The reward determination subunit is also used for: Based on the pre-trained feature extraction network, feature extraction is performed on the preset standard microstructure image to obtain the target feature vector; Based on the distance between the micro-feature vector after the current action is executed and the target feature vector, the current micro-performance deviation is determined.

17. The system according to claim 16, characterized in that The features of the microstructure image include the distribution of the reinforcement phase of the copper-tungsten electrical contact material, the microscopic morphology of the copper-tungsten electrical contact material, and the grain boundary structure of each crystal phase in the copper-tungsten electrical contact material.

18. A system for preparing copper-tungsten electrical contact material, characterized in that: include: A preparation module is used to prepare the copper-tungsten electrical contact material based on the optimal preparation parameter combination of the copper-tungsten electrical contact material obtained according to any one of claims 1-8.

19. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the method according to any one of claims 1 to 9 is implemented.

20. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, the method according to any one of claims 1 to 9 is implemented.

Images