An Intelligent Assistant for Power Plants Based on Factored MDPs

Making good operation decisions during abnormal power plant conditions represents in many cases the possibility to avoid a unit trip or having economical losses. This paper introduces AsistO, an intelligent assistant for the decision support based on decision theoretic planning techniques. It provides power plant operators with useful recommendations to (i) maintain a plant running under safe conditions, or (ii) deal with process transients when an unexpected event occurs. We present the formalism of Markov decision processes as the core of the intelligent assistant which uses a factored representation of plant states. We also show a very intutive algorithm to approximate decision models based on training data collected through random exploration routines in a simulated environment. We have tested our system in the steam generation system of a combined power plant to deal with load disturbances.