Robust Reinforcement Learning Control and Its Application Based on IQC and PSO

摘要

In this paper a novel robust reinforcement learning control based on IQC (Integral quadratic constraints) and PSO(RRLCIP) is presented, the RRLCIP utilizes a adaptive critic to estimate the decoupling performance, a neural network to generate the decoupling action, and a PI controller to control the plant after decoupling. By replacing nonlinear and time-varying aspects of a neural network with uncertainties, a robust reinforcement learning procedure results that is guaranteed to remain stable even as the neural network is being trained and solve the local minima problem, by making use of the global optimization capability of PSO, performance can be improved through the use of learning. The RRLCIP utilize a plant model to accelerate the convergence speed. Proposed RRLCIP control strategy can not only find the good performance, but also avoid of unstable behavior at learning. The simulation results for control system of collector gas pressure of coke ovens shows its validity.