一种基于NFCS形态的模糊神经网络的学习算法

A LEARNING ALGORITHM FOR FUZZY NEURAL NETWORK BASED ON NFCS FORM

神经网络与模糊逻辑协同系统 (NFCS)是神经网络与模糊系统深度融合的一种形态 .传统的 BP算法也可作为 NFCS的学习算法 ,但收敛性能不佳 .针对 NFCS形态的模糊神经网络提出了 BP算法的一种新的改进算法(NFCS- BP) ,即在误差传播时不仅改变网络的连接权值 ,同时也改变模糊逻辑神经元模型的补偿参数 .首先介绍了NFCS的协同机制和典型结构 ,然后详细推导了改进算法的迭代公式 .实践证明 ,与传统 BP算法相比 ,该算法具有收敛性能好、函数逼近精度高的优点 .

Neuron-fuzzy cooperation system (NFCS) is a form of deeply combining neural network with fuzzy system. The conventional BP algorithm can be used in NFCS too, but it is bad at convergence performance. A new improved BP algorithm (NFCS-BP) is presented in allusion to fuzzy neural network based on NFCS form, which modifies not only the weights between neurons but also the compensative parameters of fuzzy logic neuron model while error is back propagating. The principle of NFCS's cooperation and the typical construct are introduced at first, and then the iterative formula of the improved BP algorithm is derived in detail. As is pointed out by the experimented results, this algorithm is more efficient than the conventional BP algorithm on the aspects of convergence performance and the ability of function approximation.