Two approaches for automating the tuning process of fuzzy logic controllers [PWR application

Abstract

The design and evaluation by simulation of two automatically tuned fuzzy logic controllers is presented. Typically, fuzzy logic controllers are designed based on an expert's knowledge of the process. However, this approach has its limitations in the fact that the controller is hard to optimize or tune to get the desired control action. In this paper are two methods to automate the tuning process: 1) using a simplified Kalman filter approach, and 2) using a suboptimal filter. Here, the objective is to enable the fuzzy logic controller to track a suitable reference trajectory, i.e., to learn from an expert's actions. To demonstrate an application of this approach, a PWR (pressurized water reactor) type nuclear reactor model (nonlinear), based on assumed point kinetics with six-delayed neutron groups, is controlled using a fuzzy logic controller that utilizes estimated temperatures from a one-delayed neutron group observer. Here, a previously designed robust optimal controller's response is used as a reference trajectory to determine automatically the rules for the fuzzy logic controller. The fuzzy logic controller displayed good stability and performance robustness characteristics for a wide range of operation. >