Grid fuzzy models with variable boundaries and their application to monitoring of plant operations

Abstract

This paper describes a special type of fuzzy models with incomplete grid-type fuzzy rule base in the multidimensional input space. The number and locations of the fuzzy rules are automatically defined by the concrete distribution of the training data and by the assumptions for the size and the structure of the grid. A simplified iterative learning algorithm for calculating the singletons of such Grid Fuzzy Models is presented in the paper. The Grid Fuzzy Models are important element in the proposed monitoring system for plant operations. Here the moving window approach is used for model prediction based on each window data. Then a dissimilarity degree between the current operation and another, previously known one is calculated. Real plant operation data from the Stabilization unit in a petrochemical plant are used to prove that the proposed monitoring system is able to properly detect different plant operations.