Dynamic knowledge inference and learning under adaptive fuzzy Petrinet framework

摘要

Since knowledge in an expert system is vague and modifiednfrequently, expert systems are fuzzy and dynamic. It is very importantnto design a dynamic knowledge inference framework which is adjustablenaccording to knowledge variation as human cognition and thinking. Angeneralized fuzzy Petri net model, called adaptive fuzzy Petri netn(AFPN), is proposed with this object in mind. AFPN not only has thendescriptive advantages of the fuzzy Petri net, it also has learningnability like a neural network. Just as other fuzzy Petri net (FPN)nmodels, AFPN can be used for knowledge representation and reasoning, butnAFPN has one important advantage: it is suitable for dynamic knowledge,ni.e., the weights of AFPN are adjustable. Based on the AFPN transitionnfiring rule, a modified backpropagation learning algorithm is developednto assure the convergence of the weights