Continuous decision models for process control

The intelligent real-time monitoring associate (IRMA) framework for supervisory control of complex distributed production processes is presented. IRMA integrates decision analytic, optimization, and AI-style tools into a multiprocessing problem-solving environment. The application of IRMA to emissions control at a geothermal power plant is described. This application uses the Gaussian influence diagram (GID) model for control by unrolling a continuous process into a sequence of discrete-time control decisions and optimizing them, taking account of previous observations of the system state. A tutorial introduction to GIDs is provided, and approaches to model specification and repeated construction of the GIDs used for emissions control are described. Examples of the operation of the controller are given, and extensions of this method to other aspects of supervisory process control are suggested.<
>