Method of Evaluation of a Set of Linguistic Rules

摘要

The object was to improve a method which establishes how a fuzzy controller evaluates a set of linguistic rules {R1, ..., RN} at run-time and thus determines the current value of its output variable. The premise of each rule is an AND-coupling of linguistic statements about the input variables of the controller, the conclusion a linguistic statement about the output variable. The method already known provides that - in the first phase, for each of the rules, the conclusion is replaced by a counter value Zi, which is determined in that the membership function for the conclusion is "defuzzified", and - in the second phase, for each sampling interval t -- first, for i = 1,...,N, the current validity omega i (t) of the rule Ri is determined, where omega i (t) is calculated from the membership values of the current values of the input variables occurring in the premise of Ri to the membership functions which characterise the linguistic statements of the premise -- and then the current value zt of the output variable is determined according to the equation IMAGE The improvement according to the invention establishes which conditions the membership functions must satisfy for the linguistic statements about the input variables as well as the control number {R1, ..., RN}, in order that it is ensured that the divisor in the equation (1) is always 1. This avoids, at each sampling interval, a summation and a division as well as special handling for the case that the denominator in (1) is zero, and therefore the invention permits an efficient and economical realisation of the method. The conditions are that the linguistic batch of each input variable must form a fuzzy information system and that the control set is complete and consistent. Advantageous embodiments are the restriction to trapezoidal membership functions and, at each sampling interval, the pre-determination and intermediate storage of all membership values. IMAGE