Reinforcement learning for process identification, control and optimisation

This paper presents the implementation of a novel method for process modelling, control and optimisation within a reinforcement learning framework. The approach used is based on a model-free action-dependent adaptive critic design (ADAC). This is used to develop an online learning strategy to model and also to control a highly nonlinear CSTR plant. Comparison with conventional methods shows that the technique is able to achieve comparable performance without any manual intervention.