A Learning-driven Method for Adaptive PID Control

Aiming at the slow convergence speed of existing adaptive proportional-integral-derivative (PID) controllers, an adaptive PID controller driven by deep reinforcement learning is proposed. The controller uses the multi-threaded synchronous learning feature of the Advantage-Actor-Critic (A2C) structure to train multiple Actor-Critic (AC) structure agents in parallel to reduce the correlation of learning samples. Each agent adopts a radial basis function (RBF) network to simultaneously approximate the policy function and the value function to improve computational efficiency. Finally, we perform simulations in the OpenAI gym environment to illustrate the efficiency and robustness of the proposed scheme.