Abstract
The safety of chemical reactor is one of the major problems in process safety that has caused numerous incidents and has had detrimental effects on individual health and environment. The susceptibility of the Continuously Stirred Tank Reactor (CSTR) to loss of control is typically due to either human error or controller failure. In this paper, a state-of-art resilient control framework utilizing Proximal Policy Optimization (PPO) is proposed to effectively control the CSTR operation and maintain the statue at a desired steady state. The purpose of this work is to improve system resilience through implementation of resilience calculation and integration with reinforcement learning to finally reach optimal resilience. Comparative analysis is then performed to evaluate the control performance under the guidance of proposed framework compared with Proportional Integral Derivative controller (PID) and Nonlinear Model Predictive Control (NMPC). The results prove that the proposed framework is applicable to develop resilient system and considerably outperforms the PID controller and the NMPC in terms of efficiency and stability.