一种新型的动态模糊神经网络算法

Novel Dynamic Fuzzy Neural Networks Algorithm

为了解决模糊系统中的知识抽取问题和避免初值选择的任意性,提出一种新型的动态模糊神经网络算法。运用规则产生准则时,考虑输出误差和可容纳边界的有效半径这2个重要因素;通过分级学习法,大大提高学习的有效性,加之参数的调整只限于线性参数,没有迭代学习,因而学习速度很快,这使算法应用于实时学习成为可能;非线性参数是由训练样本和启发式方法直接决定的。利用D-FNN来进行Mackey-Glass混沌时间序列预测实验。仿真结果表明D-FNN算法的有效性和实用性。

In order to solve the knowledge extraction problem in fuzzy system and avoid the starting value choice haphazardness. A dynamic fuzzy neural network （D-FNN）, algorithm is proposcd. When the criterion of rule generation is used, the output error and the effective radius of the may-hold boundary are considered, as two important factors. By using the graduation study strategy＇s application, the study validity is greatly enhanced. In addition the parameter adjustment is only restricted in the linear parameter, without iterated the study. Thus the study speed is very quick. It is possible that the algorithm is applied into the real-time study. The nolinear parameter is directly decided by the training sample and the heuristic method. The Mackey-Glass chaos time series forecast experiment using D- FNN is done. The simulation result silows the effectiveness and practicability.