A Field-Path Co-operation Analysis Method and System Based on Reinforcement Learning and Bayesian Optimization

Abstract

This invention discloses a field-circuit collaborative analysis method and system based on reinforcement learning and Bayesian optimization. The invention utilizes a Bayesian optimization algorithm to optimize the parameter values ​​of each electronic component in a selected topology. Based on the optimized parameter values, a reinforcement learning-based parameter search environment for electronic components in the RF device circuit topology is constructed, modeled as a Markov decision process. An equivalent circuit model of the RF device is designed according to the optimal action. The equivalent circuit model of the RF device is connected to other circuit systems for circuit-level simulation analysis. By integrating reinforcement learning and Bayesian optimization, two intelligent algorithms, and fully leveraging their respective advantages in discrete structure search and continuous parameter optimization, a high degree of automation and intelligence is achieved throughout the entire field-circuit collaborative analysis process. While ensuring the physical rationality of the model, the efficiency and accuracy of the collaborative analysis are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic simulation and circuit system co-design, and in particular to a field-circuit co-analysis method and system based on reinforcement learning and Bayesian optimization. Background Technology

[0002] In modern electronic system design, circuit performance evaluation under high-frequency, high-speed, and complex electromagnetic environments often requires combining electromagnetic field simulation and circuit simulation, i.e., field-circuit co-analysis. Traditional methods typically model and optimize electromagnetic components and circuit components separately, resulting in low simulation efficiency, poor synergy, and difficulty in achieving optimal performance at the overall system level. The equivalent circuit model serves as a crucial bridge connecting field simulation and circuit simulation, and its design quality directly impacts the accuracy and efficiency of the co-analysis.

[0003] The equivalent circuit modeling methods commonly used in field-circuit co-operation analysis mainly fall into two categories:

[0004] 1. The characteristics of a predefined circuit topology are progressively adjusted using SPICE simulation and black-box optimization (BBO) algorithms. This method automatically ensures the passivity of the model, as the circuits consist only of passive lumped elements. Common BBO methods for optimizing lumped element parameters include Genetic Algorithm (GA), Particle Swarm Optimization (PSO), or Differential Evolution (DE). Its advantages include ensuring passivity and generating physically meaningful netlists. The disadvantages of this method are the need to manually define a suitable circuit topology, which remains fixed during optimization, potentially leading to an overly complex model. Furthermore, simulation time increases significantly with the number of circuit elements.

[0005] 2. A vector fitting (VF) method is used to generate a rational function approximation of the S-parameters, and the equivalent circuit is derived based on the fitted poles and residues. This method requires manually defining several poles to fit the rational function to the S-parameter data. Based on this fit, the circuit model can be derived. The main advantages of this method are its fast running speed and accurate model characteristics, but the resulting circuit topology often lacks physical representativeness. Furthermore, it may produce non-passive components, violating the passivity rule.

[0006] In recent years, machine learning methods have provided new approaches for equivalent circuit modeling and field-circuit co-operation analysis. Reinforcement learning can autonomously explore optimal circuit structure parameters through trial and error mechanisms; while Bayesian optimization excels at efficiently finding the global optimum in a low-dimensional parameter space, significantly improving optimization efficiency. However, existing research mostly focuses on the application of single methods and has not yet effectively combined the search capabilities of reinforcement learning with the parameter optimization advantages of Bayesian optimization to support equivalent circuit modeling and simulation in a broader sense of field-circuit co-operation.

[0007] Therefore, there is an urgent need for a collaborative analysis method that can unify the field-circuit modeling process and automatically realize structural exploration and parameter optimization, so as to improve the efficiency and reliability of system-level simulation and meet the design requirements of modern electronic systems in high-frequency, high-speed and complex integrated scenarios. Summary of the Invention

[0008] This invention aims to at least partially address one of the technical problems in related technologies. To this end, this invention proposes a field-path cooperative analysis method and system based on reinforcement learning and Bayesian optimization.

[0009] In a first aspect, the present invention provides a field-path cooperative analysis method based on reinforcement learning and Bayesian optimization, comprising the following steps:

[0010] Obtain the target frequency domain S-parameters of the RF device under electromagnetic field simulation;

[0011] The Bayesian optimization algorithm is used to optimize the parameter values ​​of each electronic component in a radio frequency device with a selected topology within a preset parameter range.

[0012] Based on the optimized parameter values ​​of each electronic component, a parameter search environment for electronic components in the circuit topology of a radio frequency device based on reinforcement learning is constructed. This environment is modeled as a Markov decision process. In this process, the reinforcement learning optimizes the strategy to enable the agent to select the optimal action in different states, thereby maximizing the expected cumulative reward.

[0013] Design the equivalent circuit model of the RF device based on the optimal action;

[0014] By connecting the equivalent circuit model of the RF device with other circuit systems, circuit-level simulation analysis can be performed, thereby achieving field-circuit co-analysis.

[0015] Preferably, after obtaining the target frequency domain S-parameters of the radio frequency device under electromagnetic field simulation, the frequency domain S-parameters need to be preprocessed.

[0016] Preferably, in the Markov decision process:

[0017] The state space is defined as the normalized error between the predicted S-parameters corresponding to the current electronic component parameter values ​​and the target frequency domain S-parameters.

[0018] Action space is defined as the discrete operation of decreasing, keeping unchanged, or increasing the parameter values ​​of electronic components in a radio frequency device circuit.

[0019] The reward function is defined based on the normalized error between the predicted S-parameters and the target frequency domain S-parameters corresponding to the current electronic component parameter values, in order to guide the agent to explore the optimal parameter values ​​of the topology structure that has both high accuracy and high physical rationality.

[0020] More preferably, the formula for calculating the normalized error in the state space is:

[0021]

[0022] in, The predicted frequency domain S-parameters corresponding to the current circuit element parameter values. The target frequency domain S-parameters.

[0023] More preferably, the reward function is defined as:

[0024]

[0025] in, The length of the frequency vector. The predicted frequency domain S-parameters corresponding to the current circuit element parameter values. The target frequency domain S-parameters.

[0026] Preferably, the Bayesian optimization algorithm needs to determine the objective function, construct a surrogate model of the objective function using a Gaussian process, construct the acquisition function of the objective function using logarithmic expectation, and select the next most promising parameter point for evaluation in order to find the optimal component parameter values.

[0027] Preferably, the optimization strategy employs a near-end strategy optimization algorithm.

[0028] Secondly, the present invention provides a field-path cooperative analysis system based on reinforcement learning and Bayesian optimization, comprising:

[0029] The data acquisition module is responsible for acquiring the target frequency domain S-parameters of the RF device under electromagnetic field simulation.

[0030] The Bayesian parameter optimization module is responsible for optimizing the parameter values ​​of each electronic component in a selected topology within a preset parameter range using the Bayesian optimization algorithm. This module includes a surrogate model manager for building and updating Gaussian process models of objective functions (such as S-parameter fitting error); a data acquisition function optimizer for calculating and selecting the next most promising parameter point for evaluation; and a circuit simulation interface for receiving parameter combinations and returning their S-parameter data.

[0031] The equivalent circuit model construction module for radio frequency (RF) devices is responsible for constructing a parameter search environment for electronic components in the circuit topology of the RF device based on reinforcement learning, using the parameter values ​​of each electronic component optimized by the Bayesian parameter optimization module. This environment is modeled as a Markov decision process. The reinforcement learning optimizes the agent to select the optimal action in different states through an optimization strategy, thereby maximizing the expected cumulative reward. The equivalent circuit model of the RF device is designed based on the optimal action.

[0032] The field-circuit co-analysis module is responsible for connecting the equivalent circuit model of RF devices with other circuit systems to perform circuit-level simulation analysis.

[0033] Thirdly, the present invention provides an electronic device including a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the field-path cooperative analysis method.

[0034] Fourthly, the present invention provides a machine-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the field-path cooperative analysis method.

[0035] The beneficial effects of this invention are as follows:

[0036] 1. High level of automation and intelligence: By autonomously exploring the circuit parameter space through reinforcement learning, the reliance on manual presets is reduced, the need for professional experience is decreased, and the degree of automation of field-circuit collaborative analysis is improved.

[0037] 2. Strong global optimization capability: Combining the exploration capability of reinforcement learning in discrete parameter space and the optimization capability of Bayesian optimization in continuous parameter space, it effectively avoids local optima and obtains a better global circuit model.

[0038] 3. Balancing efficiency and accuracy: Bayesian optimization searches for optimal parameters with high sample efficiency, reducing the number of simulations; reinforcement learning quickly learns the optimal strategy under the guidance of precise rewards, significantly shortening the design cycle while ensuring accuracy.

[0039] 4. Ensure physical rationality and passivity: The search process is limited to the space of electronic components to ensure that the final circuit naturally meets the passivity requirement and has a clear physical meaning, which is convenient for engineering applications.

[0040] 5. Highly practical and applicable: It achieves efficient collaborative simulation of field and circuit through equivalent circuit model, provides complete system implementation, integrates data processing, intelligent optimization and system-level co-simulation modules, is easy to integrate into existing EDA processes, and has high engineering practical value. Attached Figure Description

[0041] To more clearly illustrate the implementation of the present invention or the existing technical solutions, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0042] Figure 1 A flowchart of a field-path cooperative analysis method based on reinforcement learning and Bayesian optimization provided in an embodiment of the present invention;

[0043] Figure 2 A flowchart of a Bayesian optimization algorithm provided in an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of reinforcement learning provided in an embodiment of the present invention;

[0045] Figure 4 A comparison diagram of the S-parameter curves of the equivalent circuit model provided in an embodiment of the present invention and the target S-parameter curve;

[0046] Figure 5 This is a schematic diagram of the field-path cooperative analysis system structure based on reinforcement learning and Bayesian optimization provided in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0048] First, the field-path cooperative analysis method based on reinforcement learning and Bayesian optimization proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0049] Figure 1 This is a flowchart of a field-path cooperative analysis method based on reinforcement learning and Bayesian optimization according to an embodiment of the present invention.

[0050] like Figure 1 As shown, this embodiment of the invention provides a field-path cooperative analysis method based on reinforcement learning and Bayesian optimization, including the following steps:

[0051] S1: Obtain the target frequency domain S-parameter data of the RF devices in the system-level circuit under electromagnetic field simulation, and preprocess the data to provide high-quality input data for equivalent circuit modeling.

[0052] Target S-parameter data of the target RF device in the system-level circuit are obtained from electromagnetic simulation software such as HFSS. Frequency points and corresponding S-parameter values ​​are extracted and written into the input file for subsequent reinforcement learning training in a specific format.

[0053] S2: Utilize Bayesian optimization algorithms to optimize the parameter values ​​(such as resistance, capacitance, and inductance) of each electronic component in a selected topology within a preset parameter range. Bayesian optimization, by constructing a probabilistic surrogate model of the objective function (such as a Gaussian process) and a sampling function (such as log-expected improved logEI), quickly finds the optimal combination of component parameters with the fewest evaluations, greatly improving the efficiency of parameter tuning.

[0054] Furthermore, before performing Bayesian optimization, a circuit netlist input file for circuit simulation is provided. This file describes information such as the topology, port excitations, design parameters, and frequency range for frequency domain simulation of the circuit to be simulated.

[0055] Furthermore, Bayesian optimization requires determining the objective function and constructing a surrogate model and a data collection function for that objective function. The surrogate model is a statistical model used to approximate the true objective function; the data collection function defines how to select the next point for the actual evaluation of the objective function based on the predictions of the surrogate model.

[0056] Furthermore, such as Figure 2 As shown in the flowchart of the Bayesian optimization algorithm according to an embodiment of the present invention, Bayesian optimization is mainly divided into generating initial training points for the surrogate model, fitting the surrogate model with the training points, constructing a collection function based on the surrogate model and the initial collection points, and then transforming the optimization problem of the black box model into the optimization of the collection function. The optimal value of the collection function is selected as the point with the highest score and added to the training set of the surrogate model as the next collection point, thereby forming a loop of iterative optimization to obtain the optimal value.

[0057] Preferably, in this embodiment of the invention, the objective function is a black-box function, the function input is the component parameter values ​​of the selected circuit topology, and the function output is a reward function. .

[0058] Preferably, the surrogate model in this embodiment of the invention employs a Gaussian process (GP), which is capable of estimating the predicted values ​​of the function and the uncertainty of these predicted values ​​(i.e., the confidence interval of the predicted values).

[0059] Preferably, the acquisition function in this embodiment of the invention employs log-expectation improvement (logEI). This acquisition function determines the next sampling point, striking a trade-off between exploration (attempting to move away from known points to explore more areas) and utilization (selecting the best known or predicted point). Its form is:

[0060]

[0061] in, Indicates the expected value. At the sampling point The value of the random function, This is the best function value observed to date. It means only when Greater than An improvement is only considered complete when the expected improvement is very small, otherwise it is considered zero. The main purpose of taking the logarithm is to improve the robustness of numerical computation, especially in regions where the expected improvement is very small, close to zero. In such cases, directly calculating the EI may encounter numerical problems such as vanishing gradients. Taking the logarithm can avoid these problems, allowing for accurate calculation and comparison of the acquired values ​​and their gradients in these regions.

[0062] S3: Construct a circuit topology parameter search environment for radio frequency devices based on reinforcement learning. This environment is modeled as a Markov decision process (MDP). The reinforcement learning optimizes the policy to enable the agent to select the optimal action in different states, thereby maximizing the expected cumulative reward.

[0063] Furthermore, constructing a reinforcement learning environment requires determining the agent's state space, action space, and reward function. The state space represents the set of all possible states the agent can reach, the action space represents the set of actions the agent can perform, and the reward function represents the reward the agent receives after performing an action in a given state, resulting in a state transition.

[0064] It should be noted that this embodiment of the invention introduces a reinforcement learning environment, which models the design process of the equivalent circuit as a Markov decision process (MDP). The MDP process is defined as a quadruple. ,in The state space represents the state space of an agent, that is, the set of all possible states of the agent. The action space represents the action space of an agent, that is, the set of all possible actions of the agent; This represents the state transition probability, i.e., the probability of transitioning between states. Next action After transitioning to state The probability of; Indicates the state Next action The instant reward obtained afterward.

[0065] Furthermore, such as Figure 3 As shown in the schematic diagram of reinforcement learning according to an embodiment of the present invention, the basic process of reinforcement learning is as follows: the agent selects an action. When an action is performed on the environment, the environment undergoes a state transition and enters the next state after receiving the action. Selecting this action will grant you a reward. The environment feeds back the state and reward to the agent, indicating whether the agent's action is good or bad. The goal of the entire reinforcement learning process is to maximize the accumulation of rewards.

[0066] It should be noted that the determination of the agent's state space, action space, and reward function is based on the simulation environment model.

[0067] Preferably, in this embodiment of the invention, the state space includes the current circuit element parameter values ​​and the normalized error between their corresponding S-parameters and the target S-parameters. The error is calculated using the following formula:

[0068]

[0069] in, The S-parameters corresponding to the current circuit element parameter values. The target S-parameters.

[0070] In this embodiment of the invention, the action space is Where -1 represents the action of decreasing the component parameter value, 0 represents the action of keeping the component parameter value unchanged, and 2 represents the action of increasing the component parameter value.

[0071] The reward function in this embodiment of the invention is defined as follows:

[0072]

[0073] in, This represents the length of the frequency vector. If and Basically the same, Approaching zero represents the best possible reward. For larger... , The value is negative. Therefore, the reinforcement learning agent maximizes its reward. Minimize This improves exploratory behavior during training.

[0074] S4: This invention combines reinforcement learning and Bayesian optimization to design the equivalent circuit model of RF devices based on the optimal action.

[0075] Furthermore, the algorithm to be used by the reinforcement learning agent needs to be determined before collaborative optimization can be performed.

[0076] Preferably, the embodiments of the present invention employ the Proximal Policy Optimization (PPO) algorithm. The core idea of ​​PPO is to avoid excessive performance degradation by limiting the step size of policy updates. This algorithm optimizes the policy so that the agent can choose the optimal action in different states, thereby maximizing the expected cumulative reward.

[0077] Furthermore, the optimal parameters obtained through Bayesian optimization are used as the initial points for reinforcement learning training. Their reward values ​​are calculated and fed back to the RL agent for policy updates. This process is iterated until the reward converges or the maximum number of iterations is reached. The specific process is as follows:

[0078] 1. Perform Bayesian optimization on the selected circuit topology to obtain the optimal parameter values.

[0079] 2. Use the optimal parameter values ​​as the initial point for reinforcement learning training.

[0080] 3. The agent takes action according to the policy.

[0081] 4. Update the circuit netlist file.

[0082] 5. Run the simulation using the NGSPICE simulator to obtain the S-parameter data of the current circuit.

[0083] 6. Calculate the reward value and feed it back to the agent for policy updates.

[0084] 7. Repeat steps 3 to 6 until the reward converges or the maximum number of iterations is reached, at which point training ends.

[0085] After training, the final output is a passive equivalent circuit model that meets the target S-parameter requirements. For example... Figure 4 As shown, the comparison between the S-parameter curve of the equivalent circuit model output by the embodiment of the present invention and the target S-parameter curve is presented. It can be seen that the two are in very good agreement, indicating that the equivalent circuit model output by the present invention has a very good effect.

[0086] S5: Connect the generated equivalent circuit model to other circuit systems for system-level field-circuit co-analysis.

[0087] The optimized equivalent circuit model is connected to other circuits in the system-level circuitry to perform a complete system-level circuit simulation. By performing co-simulation at the system level, high-precision time-domain circuit performance analysis results are obtained, ultimately achieving efficient and accurate field-circuit co-analysis.

[0088] Next, the field-path cooperative analysis system based on reinforcement learning and Bayesian optimization proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0089] Figure 5 This is a schematic diagram of a field-path cooperative analysis system based on reinforcement learning and Bayesian optimization according to an embodiment of the present invention.

[0090] like Figure 5As shown, this embodiment of the invention provides a field-circuit cooperative analysis system based on reinforcement learning and Bayesian optimization, including: a data acquisition module, a Bayesian parameter optimization module, an equivalent circuit model construction module for radio frequency devices, and a field-circuit cooperative simulation module.

[0091] The data acquisition module is responsible for reading the S-parameter data of the target RF device from external files (such as Touchstone format .SNP files) or simulation software (such as HFSS, CST), and performing preprocessing operations such as format conversion to ensure the integrity of the input data.

[0092] The Bayesian parameter optimization module is primarily responsible for efficiently searching for optimal values ​​for components within a preset parameter range. This module includes a surrogate model manager for building and updating a Gaussian process (GP) model of the objective function (i.e., the reward function); a sampling function optimizer for selecting the next most promising sampling point based on sampling functions such as log-expectation improvement (logEI); and a parameter evaluation interface that passes the selected parameter combinations to the circuit simulation engine for simulation and receives the returned S-parameter data to calculate the reward value. This module has parallel evaluation capabilities, processing multiple parameter points simultaneously, significantly improving optimization efficiency.

[0093] The equivalent circuit model construction module for the radio frequency device includes a reinforcement learning parameter search module and a collaborative optimization control module.

[0094] The reinforcement learning parameter search module is used to execute the reinforcement learning parameter search step, and its core is an agent based on a reinforcement learning algorithm. This module has a built-in circuit action space, defining the operations allowed on the parameters of circuit topology components. In this embodiment, it allows three types of discrete actions: increasing, keeping unchanged, and decreasing passive components such as resistors, capacitors, and inductors. The state evaluator calculates the fitting error between the frequency response characteristics of the current circuit topology and the target S-parameters in real time and inputs this as state information into the agent. The agent selects actions based on the current state and gradually explores the optimal combination of circuit component parameters by interacting with the circuit simulation engine. This module supports various reinforcement learning algorithms, such as PPO and DQN, and has training and storage functions for policy networks and value networks.

[0095] It should be noted that the reinforcement learning agent in this embodiment of the invention uses the PPO algorithm, which can limit the step size of policy updates to avoid excessive performance degradation, thereby maximizing the expected cumulative reward.

[0096] The collaborative optimization control module, acting as the system's central scheduling center, is responsible for initializing the reinforcement learning environment and the Bayesian optimizer, coordinating information transfer and iterative processes between them. This module monitors the optimization process, recording key information such as parameter combinations, reward values, and convergence status for each round, and determines whether to terminate the optimization based on preset termination conditions (such as reward value convergence, maximum number of iterations, etc.). After optimization, the module outputs the final circuit netlist file and provides a visual report comparing the S-parameters of the final equivalent circuit model with the target S-parameters.

[0097] The field-circuit co-simulation module is used to connect the optimized equivalent circuit model with other circuits in the system-level circuit to perform complete system-level circuit simulation, thereby realizing the co-analysis of field and circuit at the system level and obtaining high-precision time-domain circuit performance analysis results.

[0098] In some embodiments, the field-circuit collaborative analysis system also includes a circuit simulation engine module. This module serves as a fundamental support module for the system, integrating an NGSPICE simulator. It is responsible for receiving the circuit netlist file generated by the aforementioned module, performing frequency domain simulation, calculating the S-parameters of the current circuit topology within a specified frequency range, and returning the results to the calling module. This module features simulation result caching and error handling functions to ensure the stability and efficiency of the simulation process.

[0099] It should be noted that the SP analysis of the NGSPICE simulator is used in the embodiments of the present invention, which can directly obtain the S-parameter data of the input circuit netlist file, greatly shortening the simulation time.

[0100] The proposed field-circuit co-simulation method and system based on reinforcement learning and Bayesian optimization integrates these two intelligent algorithms, leveraging their respective strengths in discrete structure search and continuous parameter optimization. This achieves a high degree of automation and intelligence throughout the entire field-circuit co-simulation process, significantly improving the efficiency and accuracy of the co-simulation while ensuring the physical rationality of the model. The system possesses excellent scalability and practicality, and can be integrated into existing EDA toolchains (such as ADS) to support complex system-level electromagnetic circuit co-simulation tasks. It is widely applicable to equivalent circuit modeling tasks in field-circuit co-simulation in fields such as RF circuits and microwave devices, demonstrating significant engineering application value and market potential.

Claims

1. A field-path cooperative analysis method based on reinforcement learning and Bayesian optimization, characterized in that, Includes the following steps: Obtain the target frequency domain S-parameters of the RF device under electromagnetic field simulation; The Bayesian optimization algorithm is used to optimize the parameter values ​​of each electronic component in a radio frequency device with a selected topology within a preset parameter range. Based on the optimized parameter values ​​of each electronic component, a parameter search environment for electronic components in the circuit topology of a radio frequency device based on reinforcement learning is constructed. This environment is modeled as a Markov decision process. In this process, the reinforcement learning optimizes the strategy to enable the agent to select the optimal action in different states, thereby maximizing the expected cumulative reward. Design the equivalent circuit model of the RF device based on the optimal action; By connecting the equivalent circuit model of the RF device with other circuit systems, circuit-level simulation analysis can be performed, thereby achieving field-circuit co-analysis.

2. The method according to claim 1, characterized in that, After obtaining the target frequency domain S-parameters of the RF device under electromagnetic field simulation, the frequency domain S-parameters need to be preprocessed.

3. The method according to claim 1, characterized in that, The state space of the Markov decision process is defined as the normalized error between the predicted S-parameters and the target frequency domain S-parameters corresponding to the current electronic component parameter values; the action space is defined as performing discrete operations to decrease, remain unchanged, or increase the electronic component parameter values ​​in the radio frequency device circuit; the reward function is defined based on the normalized error between the predicted S-parameters and the target frequency domain S-parameters corresponding to the current electronic component parameter values, in order to guide the agent to explore the optimal parameter values ​​of the topology.

4. The method according to claim 3, characterized in that, The formula for calculating the normalized error in the state space is: ； in, The predicted frequency domain S-parameters corresponding to the current circuit element parameter values. The target frequency domain S-parameters.

5. The method according to claim 3, characterized in that, The reward function is defined as follows: ； in, The length of the frequency vector. The predicted frequency domain S-parameters corresponding to the current circuit element parameter values. The target frequency domain S-parameters.

6. The method according to claim 1, characterized in that, The Bayesian optimization algorithm requires determining the objective function, constructing a surrogate model of the objective function using a Gaussian process, improving the objective function using logarithmic expectation, and selecting the next most promising parameter point for evaluation in order to find the optimal component parameter values.

7. The method according to claim 1, characterized in that, The optimization strategy employs a near-end strategy optimization algorithm.

8. A field-path cooperative analysis system for implementing the method of claim 1, characterized in that, include: The data acquisition module is responsible for acquiring the target frequency domain S-parameters of the RF device under electromagnetic field simulation. The Bayesian parameter optimization module is responsible for optimizing the parameter values ​​of each electronic component in a selected topology within a preset parameter range using the Bayesian optimization algorithm. The equivalent circuit model construction module for radio frequency (RF) devices is responsible for constructing a parameter search environment for electronic components in the circuit topology of the RF device based on reinforcement learning, using the parameter values ​​of each electronic component optimized by the Bayesian parameter optimization module. This environment is modeled as a Markov decision process. The reinforcement learning optimizes the agent to select the optimal action in different states through an optimization strategy, thereby maximizing the expected cumulative reward. The equivalent circuit model of the RF device is designed based on the optimal action. The field-circuit co-analysis module is responsible for connecting the equivalent circuit model of RF devices with other circuit systems to perform circuit-level simulation analysis.

9. An electronic device, characterized in that, The method includes a processor and a memory, the memory storing machine-executable instructions that can be executed by the processor, the processor executing the machine-executable instructions to implement the method of any one of claims 1-8.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores machine-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method described in any one of claims 1-8.

Images