Decoupling Control Using a PSO-Based Reinforcement Learning

In this paper an intelligent decoupling control architecture using evolutionary reinforcement learning (IDCERL) is presented. The IDCERL utilizes an adaptive critic to estimate the decoupling performance, and a TSK fuzzy neural network (TSFN) to generate the decoupling action. By making use of the global optimization capability of particle swarm optimization (PSO), the IDCERL can solve the local minima problem in traditional actor-critic reinforcement learning. The IDCERL utilize a plant model to accelerate the convergence speed and void the possible risks of large disturbance of action generated by PSO global search. Proposed control strategy can reduce the controller developing time by incorporating prior knowledge in a fuzzy neural network form. The application for control system of collector gas pressure of coke ovens shows its validity.