Prof. Jay H. Lee from our KAIST CBE Department gave a plenary presentation at the 13th International Symposium on Process Systems Engineering (PSE 2018), which is a triennial conference attracting more than 500 attendees and was held in San Diego, CA, this year. The talk and the paper were prepared and delivered in collaboration with Dr. Thomas Badgwell at ExxonMobil Research and Engineering Company, USA. The presentation started with a brief introduction to Reinforcement Learning (RL) technology, summarizing recent developments in this area, and discussed their potential implications for the field of process control and beyond. First, a brief introduction was given on RL, a machine learning technology that allows an agent to learn, through trial and error, the best way to accomplish a task. Then two new developments in RL were discussed that have led to the recent wave of applications and media interest. A comparison of the key features of RL and Model Predictive Control (MPC) was then presented in order to clarify their similarities and differences. This was followed by an assessment of five ways that RL technology can potentially be used in process control applications. Then, applications of RL to process systems engineering (PSE) problems beyond control were outlined. Specifically, the problem of multi-scale sequential planning/operation decision under stochastic uncertainty and its complications were addressed and the potential of using RL combined with mathematical programming was suggested as a way to tackle this previously intractable problem.
Keywords : Reinforcement Learning, Model Predictive Control, Process Control. Multi-scale Sequential Decision Under Uncertainty
