An Evolutionary Approach for Fuzzy Knowledge Learning

Adaptive Fuzzy Petri Nets (AFPN) were proposed for knowledge reasoning and learning. They have advantage on learning dynamical knowledge, i.e., weights of an AFPN model are adjustable dynamically according to knowledge update. In this paper, an evolutionary algorithm called Adaptive Weights Evolutionary Algorithm (AWEA) is introduced which is capable of guaranteeing convergence of AFPN weights. Simulation results show effectiveness of AWEA. Comparing with the original back propagation learning algorithm of AFPN, AWEA does not depend on initial parameters to achieve convergence, so it avoids of getting trapped in local minimum. Additionally, AWEA converges faster than Back propagation algorithms.