AI planning in supervisory control systems

This paper presents an AI planning-based framework to support the activities of a human operator in a supervisory control system. The framework uses an AI planning and learning substrate architecture and is designed for integration within a general third-party real time software. Our goal is to build a test-bed architecture for research in AI planning applications, such as electrical and industrial processes. AI planning techniques, as opposed to the more traditionally used rule-based systems, can be useful in the automation of the supervision of process systems, as they provide rich planning representations and algorithms. We present our work on developing an AI planning system for a boiler power plant domain. We develop a set of planning operators from an extended multilevel flow modeling (MFM) of the plant. Our planner reasons about goals and its subgoals, generates plans for different scenarios, including the sequence for start-up of the plant. We show our approach to acquire the domain knowledge, which is a well-known difficult enterprise for real-world applications. We demonstrate that our modeling approach built upon MFM, is successful in mapping the supervisory system knowledge into a planning representation