From adaptive to evolving fuzzy control policy: Critical processing steps and configuration

摘要

In previous work, the efficiency of a new Adaptive Fuzzy Logic Controller (AFLC) was successfully tested on a real time control process. It's to know that the adaptation process do not require any kind of offline pre-training or information about the mathematical model of the plant. Starting from a defined structure of Fuzzy Logic Controller, the error and its derivative were the only information used to tune this adaptive controller. The adaptation and self learning are done by adjusting their internal parameters in real time. These papers aim to expose and focus on the enhancement proposed by the original authors of this control algorithm in order to proceed to a series of comparisons using two different control processes (Tank level and temperature of a room) that can reflects the control performances improvement. The Adaptive controller performances are improved by giving him more freedom and endowing him by evolved tools that can strongly support any kind of unexpected changes on the controlled plant behaviors. The controller exposed here can start the control with a very basic structure, assigning (arbitrarily) one membership function for each input. Through two stages, the control algorithm starts the adaptation and the control of the plant firstly by adjusting the rules consequences using the output error then in the second stage and when the adaptation process cannot bring no more improvement, the controller starts the evolved process to get a new topology that can guarantee more accuracy by adding new membership functions in the inputs spaces.