Deep Reinforcement Learning Control of a Boiling Water Reactor
Xiangyi Chen, Asok Ray
Abstract
This article presents (nonlinear) control system synthesis for a boiling water reactor (BWR) by using artificial intelligence (AI)-based reinforcement learning (RL), where the pertinent algorithm is deep deterministic policy gradient (DDPG). The BWR model, used in this article, exhibits limit cycling and/or chaotic behavior in different regions of operation. The performance of the RL control system is compared with that of a control system synthesized by the standard <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {H}_\infty $ </tex-math></inline-formula> theory. The results of comparison show that the RL control system outperforms the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {H}_\infty $ </tex-math></inline-formula> control system for disturbance rejection, stability under perturbation, and set-point tracking in a majority of the test cases.