NEURO-FUZZY MODELING AND CONTROL OF A BATCH PROCESS INVOLVING SIMULTANEOUS REACTION AND DISTILLATION

This paper is concerned with a novel approach to batch process automat ion using fuzzy modelling and reinforcement learning. The core part of the automation strategy is an autonomous agent that continuously lear ns to implement control actions that can drive the batch process' stat e very close to the desired one with near-optimal performance. An effi cient algorithm for reinforcement learning called fuzzy Q-learning is proposed to build the agent (controller). The use of linguistic inform ation to guide the learning process and to implement near-optimal acti ons provides the means for both knowledge integration and scaling rein forcement learning. The methodology is exemplified using a batch proce ss involving simultaneous reaction and distillation.