System Identification for Rule-Based Systems
Rule-based fuzzy systems translate qualitative, vague and imprecisely formulated human experience and judgment into control rules. The first section presents the technique to capture this experience. As has been said in Chapter 2, it is the human operator that is being identified while he is controlling the plant/process, thus the rules of the fuzzy control algorithm will constitute an inverse of the plant/process input-output relationship. Membership functions are determined empirically, on the basis of given guidelines, by trial-and-error. In another case, the control system designer himself is called upon to act as a human operator and formulate the rules on the basis of his engineering judgment rather than previous experience with the system. He still uses fuzzy rules and membership functions and the inverse aspects of human control still prevail. As before, membership functions are determined empirically, on the basis of given guidelines, or by trial-and-error. A further variety of the previous methods occur when the variables are inherently fuzzy, i.e. cannot be quantified in any way. This aspect affects mostly the choice of membership functions which, in this case, carry the bulk of the empirical knowledge of the human operator or designer, while the rules are established as before. In addition, a brief overview of methods that avoid the use of human operator interviews and “automatically” generate rules and/or membership functions are presented for the case when no experienced human operators are available. In this case, the rules and/or membership functions are learned from on-line measurements.
KeywordsMembership Function Fuzzy Rule Fuzzy Model Fuzzy Control Fuzzy Controller
Unable to display preview. Download preview PDF.