Optimization of intrinsically safe circuits based on Reinforcement Learning

This paper is mainly devoted to solving the parameter optimization problem of intrinsically safe composite circuits. In the design process of intrinsically safe circuits, it is often necessary to solve a series of parameter optimization problems. When the topology is relatively fixed, the optimization problem under most parameters is solved by training a model for different parameters. Unlike the traditional optimization method, different parameters are sampled and trained independently, but through reinforcement learning, The parameter optimization problem is abstracted as environment and agent, the optimization results are obtained by strategy network, the quality of strategy is evaluated by value network, the parameter optimization problem involved in intrinsically safe circuit is solved, and the training process is accelerated; Using the generalization ability of reinforcement learning, we can quickly get the optimal solution in the new intrinsically safe circuit optimization problem, and solve the intrinsically safe circuit optimization problem under different optimization objectives.