Design of stability-guaranteed neurofuzzy logic controller for nuclear steam generators

Abstract

A neurofuzzy logic controller (NFLC), which is implemented by using a multilayer neural network with special types of fuzzifier, inference engine and defuzzifier, is applied to the water level control of a nuclear steam generator (SG). This type of NFLC has the structural advantage that arbitrary two-input, single-output linear controllers can be adequately mapped into a set of specific control rules of the NFLC. In order to design a stability-guaranteed NFLC, the stable sector of the given linear gain is obtained from Lyapunov's stability criteria. Then this sector is mapped into two linear rule tables that are used as the limits of NFLC control rules. The automatic generation of NFLC rule tables is accomplished by using the back-error-propagation (BEP) algorithm. There are two separate paths for the error back propagation in the SG. One considers the level dynamics depending on the rank capacity and the other takes into account the reverse dynamics of the SG. The amounts of error back propagated through these paths show opposite effects in the BEP algorithm from each other for the swell-shrink phenomenon. Through computer simulation it is found that the BEP algorithm adequately generates NFLC rule tables according to given learning parameters.