Energy System Diagnosis by an Expert System with a Genetically Evolved Database of Fuzzy Rules

The problem of malfunction diagnosis in energy systems can be approached using an expert system which compares the experimental data measured by the plant acquisition system and the calculated data evaluated by a plant simulator under the same operating conditions. In this paper the rules that form the "knowledge base" of the expert system are not assigned heuristically by trying to code the expertise of plant personnel, as it is usually done, but they are artificially and randomly generated by the recombination and selection operators of an evolutionary algorithm. A two-objective optimization problem is set up, in order to search for the optimal sets of rules having the minimum complexity but simultaneously maximizing the number of correct fault identifications for a given set of malfunctioning operating conditions. A global and a local approach are applied to a real test case, a two-shaft gas turbine used as the gas section of a combined-cycle cogeneration plant, in order to evaluate the potentialities and the limits of this methodology.