(179a) Reinforcement Learning (RL) Based Data-Driven Decision Support System | AIChE

(179a) Reinforcement Learning (RL) Based Data-Driven Decision Support System

Authors 

Goel, P. - Presenter, Texas A&M University
Pistikopoulos, E., Texas A&M Energy Institute, Texas A&M University
Mannan, M. S., Texas A&M University
Datta, A., Texas A&M University
Big data has transformed from a buzzword to a real value creator. In recent years there has been a rising interest in data analytics in the energy industry. The industry and academia have made significant progress in big data analytics. However, there is still a need to use the plant data to gain insights and address the issues related to abnormal situation management. An abnormal situation is defined as “a disturbance or series of disturbances in a process that causes plant operations to deviate from their normal operating state.” The disturbance can be a minor near-miss to a catastrophic incident which may cause production losses, environmental damage, injuries or sometimes fatalities if not managed properly. During an abnormal situation the operator is expected to take appropriate action to bring back the process to a normal state. Diagnose and mitigation actions depend on the operator’s experience, process complexity and stress levels of the operator.

This paper outlines a systematic framework for application of Reinforcement Learning (RL) during process operations and abnormal situation management. A novel method using Reinforcement Learning is established to address decision making challenges and advise operators during abnormal situations. The system is designed on the expert knowledge from operators and engineers, the knowledge available from previous operation experiences and process plant information. The application of the proposed method can be expanded by connecting the system to Operator Training Simulators (OTS) and train operators for abnormal situation management.