Factorial Kernel Dynamic Policy Programming for Vinyl Acetate Monomer Plant Model Control

This research focuses on applying reinforcement learning towards chemical plant control problems in order to optimize production while maintaining plant stability without requiring knowledge of the plant models. Since a typical chemical plant has a large number of sensors and actuators, the control problem of such a plant can be formulated as a Markov decision process involving high-dimensional state and a huge number of actions that might be difficult to solve by previous methods due to computational complexity and sample insufficiency. To overcome these issues, we propose a new reinforcement learning method, Factorial Kernel Dynamic Policy Programming, that employs 1) a factorial policy model and 2) a factor-wise kernel-based smooth policy update by regularization with the Kullback-Leibler divergence between the current and updated policies. To validate its effectiveness, FKDPP is evaluated via the Vinyl Acetate Monomer plant (VAM) model, a popular benchmark chemical plant control problem. Compared with previous methods that cannot directly process a huge number of actions, our proposed method leverages the same number of training samples and achieves a better control strategy for VAM yield, quality, and plant stability.