基于BP网络的模糊Petri网的学习能力

模糊Petri网(Fuzzy Petri Nets，FPN)是基于模糊产生式规则的知识库系统的良好建模工具，但自学习能力差是模糊系统本身的一个缺点．该文提出了适合模糊Petri网模型自学习的模糊推理算法和学习算法．在模糊推理算法中，通过对没有回路的FPN模型结构进行层次式划分以及建立变迁点燃和模糊推理的近似连续函数，从而把神经网络中的BP网络算法自然地引入到FPN模型中．在FPN模型上，用误差反传算法计算一阶梯度的方法对模糊产生式规则中的参数进行学习和训练．经过学习和训练的FPN具有很强的泛化能力和自适应功能．FPN模型经过训练得到的参数是有特定含义的，可以通过对这些参数的合法性分析，使得模糊产生式规则系统更加有效，也对知识库系统的建立、更新和维护有着重要的意义．

Learning Capability in Fuzzy Petri Nets Based on BP Net

Fuzzy Petri Nets(FPN)are a powerful modeling tool for knowledge -based systems based on fuzzy production rule. But the lack of learning mechanism is the weakness of fuzzy systems. In this paper,a fuzzy reasoning algorithm and a learning algorithm are presented. In the fuzzy reasoning algorithm, the structure of FPN modeling without loop is partitioned into several levels, FPN is transformed into hierarchy model, at the same time, several continuous functions are built to approximate transition firing and fuzzy reasoning. Compared with other fuzzy reasoning algorithm, it is suitable to FPN learning.So the back propagation algorithm of neural networks is introduced into FPN naturally. According to calculating first derivative on error back propagation, the parameters of fuzzy production rules can be learned and trained in FPN. Data gained from FPN learning have unambiguous meaning. Analyzing these data, fuzzy production rule systems can be constructed effectively. FPN without loop can be transformed into hierarchy model, it is proved in this paper. These techniques in this paper are quite general and can be applied to most practical Petri net models and fuzzy systems.