Noise Parameterization of Continuous Deep Reinforcement Learning for a Class of Non-linear System

Abstract

Reinforcement learning (RL) is one of the most important algorithms for artificial intelligence. DDPG as continuous controller approach which can work at continuous and high dimensional data applies in this paper to solve nonlinear valve system. The aim of this paper is gaining analysis of the DDPG noise parameterization. Noise parameter addition is known to be able to increase the exploration ability of the algorithm. The noise parameterization is using the Ornstein-Uhlenbeck (OU) noise injection. This exploration investigation concerns to the algorithm’s performance. The evaluation measurement is based on the total reward to system during training. The result indicates that noise parameterization affects the performance of the algorithm. The comparisons show that the injection of OU noise for DDPG algorithm influences the total reward. The simulation find that the total reward that is achieved by DDPG with OU noise injection is higher than DDPG without OU noise injection at 317,810.