A Detailed Approach to Reinforcement Learning: a Semi-Batch Reactor Case Study

No Thumbnail Available
Date
2015-12-28
Authors
Mustafa, Mustafa Abbas
Wilson, Tony
Journal Title
Journal ISSN
Volume Title
Publisher
UOFK
Abstract
The transient nature of semi-batch reactors, coupled with the unavailability of accurate mathematical models and online measurements, restricts achieving optimal operation. However, one finds that operators have managed, through experience, to improve on previous performance. Reinforcement Learning (RL) has already been identified as an approach to mimic this interactive learning process. Core elements have not been presented in detail for direct application. This work aims to provide a blueprint of the RL approach and a validation, through MATLAB implementation, against a published case study. Moreover, the initial training data set is modified to confirm the convergence of the algorithm.
Description
Keywords
Reinforcement Learning, ANN, Optimization, Control, Semi-Batch Reactor
Citation