Synthesis of insightful algorithms

Abstract

A methodology is developed for the synthesis of Insightful Algorithms. We define "Insightful Algorithm" as an analog of a biological system which is capable of fusing environmental observations with multiple predictive models of its process and/or its relationship to its environment, in the form of a Generalized Body of Evidence, which it then uses to respond to contingencies. These combine probability density functions arising from the environmental observations and the sensors measurements, to membership functions representing the expectations of the system's models. This couples Probability Theory to Fuzzy Set Theory, and uses Zadeh's concept of the probability of a Fuzzy event. Estimation of the future performance level uses a methodology based on the Algebra of Logic and Measure Theory where exact analytical representations of the fuzzy logical operations are derived.

The implementation of the constructed system is achieved by using the Blackboard Paradigm. In this context, it acts as a synchronization scheme over a distributed processing configuration, a global database, and a job control language. The multiple processors are used for distributing the tasks over the network among different processors with the job control file on a higher hierarchy. A symbolic model based on the Emergency Operating Procedures (EOPs) is used in the form of production rules describing logically the system's operational mode. The system's models chosen are based on the statistical forecasting and time series analysis methods.

The developed methodology is applied to the task of monitoring a nuclear power plant with the objective of maximizing its future performance and anticipating any undesirable future conditions so as to steer the system away from them rather than just react to them. The synthesis of the modeling approach is illustrated for the construction of an insightful nuclear plant monitoring system for a Boiling Water Reactor. The insightful ability of the proposed approach is compared to results of actual nuclear power plant transients, to transients generated by numerical thermal hydraulic codes, and to transients generated on plant simulators. It appears that the proposed methodology offers a promising approach to the task of the predictive monitoring a nuclear power plant.