An Approach to Tune PID Fuzzy Logic Controllers Based on Reinforcement Learning
Hacene Rezine, Rabah Louali, Jrome Faucher, Pascal Maussio · InTech eBooks · 2008
On the whole, this paper provides some new and original headlines for the on-site tuning PID like fuzzy logic controllers.With the Broida methodology for conventional PID controllers; it is possible to obtain satisfying basic FPID. These settings depend only on two or three parameters (K, T and eventually ) extracting from an open-loop identification test of the process. A second set of pre-defined settings for the controllable factors FPID is proposed and selected from the reinforcement learning design procedure. Two discrete tuning algorithms are developed based FQL for antecedent parameters and consequent parameters tuning respectively. The process performances are always evaluated considering the IAE-criterion between the reference and the measured signal. The FPID will then be all the more robust since this IAE-criterion remains insensitive to uncontrollable factors (here white noise, process misidentification or high order process). This second stage tuning could be seen as an on-site roughly FPID tuning strategy.