Critic Learning Control Based on Particle Swarm Optimization for Constrained Nonlinear Systems

For nonlinear systems with asymmetric control constraints, this paper designs a critic learning control algorithm based on the particle swarm optimization (PSO). The proposed algorithm skillfully utilizes the PSO method to execute the policy improvement step, not only effectively addressing the issue of asymmetric constraints on control inputs but also eliminating the need for complex utility functions, while ensuring the optimality of the control policy. Compared to the traditional policy improvement that rely on gradient information, the proposed algorithm remains applicable even when gradient information of the system dynamics is unavailable. Furthermore, to reduce the approximation error, this paper adopts the radial basis function neural network to approximate the value function. Finally, the effectiveness of the proposed algorithm is verified through a simulation experiment.