基于 Hopfield 网络的模糊神经网络优化方法

Fuzzy neural network optimization method based on Hopfield networks

针对训练模糊神经网络时收敛时间慢，难以实时实现的缺点，将Ｈｏｐｆｉｅｌｄ网络引入模糊神经网络权值的优化问题中，从而融合了Ｈｏｐｆｉｅｌｄ网络可进行实时优化和模糊神经网络可引入专家知识化权值的优点。理论分析和模拟实验表明，这种网络可以在电路时间常数数量级内给出优化后的模糊神经网络权值，并且具有Ｌｙａｐｕｎｏｖ意义下的稳定性，为模糊神经网络权值的实时优化提供了一条新途径。

The traditional training methods of fuzzy neural networks (FNN) have disadvantages of slow convergence. This paper introduces the Hopfield networks into the optimization of weights of FNN. This method combines the advantages of real time optimization. Theoretical analysis and simulation show that this method can converge within circuit times, at the mean time, its asymptsti stable in the Lyapunov meaning and we also show that its stable. This method also provides another way for real time optimization of FNN.