模糊神经Petri网算法优化及其收敛性分析

针对模糊神经Petri网(fuzzy neural Petri net,FNPN)学习算法计算精度低、收敛性差及训练过程中网络震荡较大的问题,提出一种优化的FNPN算法.本算法采用两种S型连续函数分别表示变迁使能及变迁点燃后的新标记值,并在传统参数修正公式的基础上考虑修正前的数值进而增加新型的动量项,从而改善网络的收敛性.本文证明了优化后的参数修正算法可以保证FNPN网络的收敛性.

Optimization of Fuzzy Neural Petri Net Algorithm and Convergence Analysis

Aimed at the poor computational accuracy and convergence as well as too violent network concussion during the training of fuzzy neural Petri net （FNPN） learning algorithm, an optimized algorithm was proposed. Two S-type continuous functions were used to express transition enablement and the new tag values after transition firing. In addition, the value before correction was considered, and then the new momentum was added based on the traditional parameter correction formula, which can ensure the convergence of the proposed optimization algorithm. It was proved that the optimized parameter correction algorithm can ensure the convergence of the FNPN network.