Successive identification of a fuzzy model and its applications to prediction of a complex system (Q1177689)

The studied model is described as a series of ``if-then'' statements: --- if \(x_ 1\) is \(A_ 1^ i\) and \dots and \(x_ n\) is \(A_ n^ i\) then \(y_ i=c_ 0^ i+\sum_{j=1}^ n c_ j^ ix_ j\), \(i=1,2,\dots,K\), where the condition part (premise) involves fuzzy sets defined in spaces of values of the input variables \(x_ i\) while the action part (consequent) includes a linear combination of them. Each of the above statements are viewed as ``local'' linear regression models. Their scope of interest is focussed by the fuzzy sets standing in the condition part of the rule. The paper deals with a detailed algorithm of successive identification of the above model. The identification procedure is split into two levels, namely a supervisor level and an adjustment one. The first one determines a global policy regarding parameter adjustments while the latter executes it. The parameter adjustment refers both to the premise and consequent part of the rules and is carried out by the weighted recursive least square method.