Research And Analysis Of Parameters Optimization Of Fuzzy Petri Nets

Fuzzy Petri Nets (FPN) are a powerful modeling tool for knowledge-based systems based on fuzzy production rules.But lack of learning mechanism is an unnegligible weakness of fuzzy systems.Some parameters in fuzzy production rules,which are usually dependent upon experts'experience,are very difficult to be determined accurately.This hinders knowledge reasoning and generalizing capability of fuzzy petri nets.How to combine learning capability of neural network with fuzzy petri nets,which on one hand could make FPN possess some generalizing capability after learning and training,on the other hand overcome dependence upon experience when constructing the FPN model and let parameters adjust better the fact instance.This is the main problem to be researched in the thesis.Moreover,how to simplify a fact FPN model and make it adjust better the requirements of factual application,is the other problem to be discussed.In this thesis,based on discussing the procedure of simplifying a FPN model and analyzing its fuzzy reasoning mechnism,four algorithms in the AI area—GA,BP,ACA and CSA were introduced into the procedure of exploring parameters of FPN.Concerned with features and performance,comparison and analyses of these algorithms were presented in terms of an example on a knowledge-based system.Simulation experiment showed that ant colony algorithm (ACA) among them possessed optimal performance on the whole and the parameters trained from ACA were highly accurate.The final FPN model learned and trained from ACA was provided with strong generalizing capability and self-adjusting purpose.