Research On Adaptive Ability Of Fuzzy Petri Net Based On Improved Harmony Search Algorithm

Fuzzy Petri net(FPN),a backward-extension high level Petri net(PN)formalism,is widely used to describe the structure of fuzzy production rule(FPR)in the knowledge base system by combining PN with fuzzy theory.The three types of parameters(weights,thresholds,and certainty)of FPN are always determined by the pointed expert experience.Hence,the FPN exists some drawbacks such as poor self-learning adjustment ability and weak adaptive ability.How to improve the adaptive ability of FPN using the swarm intelligence optimization algorithm has become one of the hotspots in the FPN domain.This thesis proposes an improved harmony search algorithm by considering the characteristics of FPN to optimize the corresponding parameters for further enhancing the adaptive ability of the FPN.To sum up,the main contributions of this thesis could be divided into two aspects.(1)A hybrid harmony search and artificial bee colony algorithm with levy flight(HS-ABC-LF)is proposed to solve the inherent shortcomings of the traditional harmony algorithm,such as slow convergence speed and low search accuracy.The details of modification could be summarized into following three points.To begin with,a linear adjustment method is realized using an improved adaptive equation in the new harmony vector generation stage.Then,artificial bee colony operators are used to enhance the convergence and accuracy of the algorithm in the improvisation stage;Thirdly,the Levy flight mechanism is introduced to increase the diversity of the harmony vector,expand the search range,and jump out of the local optimal value faster in the update phase.Furthermore,the proposed HS-ABC-LF algorithm and the other three HS variants are used to test through 10 classic test functions for highlighting the effectiveness of the improved algorithm.Experimental results reveal that the proposed HS-ABC-LF has better global search capabilities and faster convergence speed.(2)The proposed HS-ABC-LF algorithm is applied to the parameter optimization process to enhance the adaptive ability of FPN.The evolution process comparison results and the fitness curves of the HS-ABC-LF algorithm are given and discussed in detail.The 20 randomly generated non-sample input data are used to perform the inference process on the FPN,which is optimized by different HS variants,for verifying the robustness of the proposed approach.The experimental results show that the FPN model optimized by the HS-ABC-LF algorithm has stronger adaptive ability and higher accuracy of inference results.