An Intelligent Assistant for Power Plant Operation and Training Based on Decision-Theoretic Planning

An Intelligent Assistant for Power Plant Operation and Training Based on Decision-Theoretic Planning

Alberto Reyes (Instituto de Investigaciones Eléctricas, México) and Francisco Elizalde (Instituto de Investigaciones Eléctricas, México)
DOI: 10.4018/978-1-60960-165-2.ch012
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In this chapter we present AsistO, a simulation-based intelligent assistant for power plant operators that provides on-line guidance in the form of ordered recommendations. These recommendations are generated using the formalism of Markov decision processes over an approximated factored representation of the plant. The decision model approximation is based on machine learning tools. We also described an explanation mechanism over these recommendations based on i) the selection of a relevant variable and ii) the automated construction of graphical explanations for operators. The explanation module analyzes the recommender system’s decision model to support the reason why a recommendation should be followed.
Chapter Preview
Top

Introduction

One of the essential requirements in the operation of complex processes, such as power plants, is to deal with a great amount of information for the decision making process. For instance, under unusual or abnormal situations, an operator must select the relevant information to rapidly identify the source of the problem, and effectively define the action plan to set the process into safe conditions. It is under these circumstances that the support of an intelligent assistant system (IAS), capable to interact with the operator and carry out some support functions required by the system, can become an important aid.

Intelligent assistants (IA) are knowledge-based systems for the decision support that provide users with accurate information at the right moment, and do suggestions and critiques during the decision making process. Optionally, an assistant could also adapt to users through simply observing their general behavior. In other words, the system could learn from users’ experiences. Among the most representative work in the field of intelligent assistants applied to industry, ASTRAL (Caimi, Lanza, & Ruiz-Ruiz, 1999) is a simulator-based assistant for power operator’s training. Its main functions are the recognition of the actions executed by an operator in a plant simulator, and the classification of detected errors with respect to an expected behavior. The main effort in this work is oriented to the development of explanation systems that support the operator’s understanding about the plant state. SOCRATES (Vale et al., 1998) is a real time assistant for control center operators in alarm processing and energy restoration. The core of the system is SPARSE, an expert system initially developed to use it in power transmission and distribution control centers. SOCRATES also provides an intelligent tutor, SPARSE-IT, which fulfills two purposes: i) to show users how a trained operator solves problems and ii) to train the user to deal with specific situations and evaluate its performance. SART (Brézillion, Gentile, Saker, & Secron, 1997) is a traffic control support system for the French subway system, and implements an intelligent tutor with several functions such as: knowledge acquisition from operators, traffic simulation, model management of the network to test with alternative cases, enumeration of alternative solutions to incidents, and training of new operators.

In this chapter we present AsistO, a simulation-based intelligent assistant for power plant operators that provides on-line guidance in the form of ordered recommendations. It includes an explanation mechanism over these recommendations to support the reason why a recommendation should be followed. The addition of this feature will allow the system to be an assistant not only for the decision support but also as an aid for operators training. As other authors, we consider the problem of generating recommedations as a sequential optimization problem that can be modeled as a Markov decision process (MDP). For instance, (Gui, David, & Ronen, 2002) describe a particular MDP model, its initialization using a predictive model, the solution and update algorithm, and its actual performance on a commercial site. They also describe the particular predictive model they used which outperforms previous models. Their system is one of a small number of commercially deployed recommender systems. The work reported in (Reyes, Sucar, & Ibarguengoytia, 2005) was a pionneer in modelling an industrial problem as a Markov decision process in which the action commands obtained help users to maintain a safe operation in the steam drum of a small power plant process. More recently, (Taghipour, Kardan, & Ghidary, 2007) proposed a reinforcement learning approach and a framework in which a web system interacts with the user and learns from its behavior.

Complete Chapter List

Search this Book:
Reset