摘要
在自组织模糊神经网络(SOFNN)算法的基础上提出了一种基于熵判据的改进算法。依据动态自适应方式建立模糊神经网络,采用误差均方根判据和误差熵判据相结合的修剪策略,对网络进行剪裁,去掉对网络输出贡献小的节点。算法的主要优点在于:能够自动地决定神经模型的结构并得出模型的参数,而不需要对神经网络和模糊系统有深入的理论知识,算法具有非常高的预测精度,并且通过修剪策略提高网络的泛化能力。应用该算法对典型的混沌时间序列Mackey-Glass序列进行了研究,结果表明,应用新的修剪策略后,算法精度及泛化能力进一步提高,并且需要的先验知识少,更适合于实际应用。
An improved algorithm with entropy criterion based on the Self-Organizing Fuzzy Neural Networks was presented. The fuzzy neural networks can be structured with self-organizing method. The less important neural nodes are pruned according to the pruning strategy with error criterion and entropy criterion. The main advantages of this approach are as followings. Firstly, it is can automatically determine the model structure and identify the model pa- rameters without requiring the in-depth knowledge about fuzzy systems and neural networks. Secondly, it provides the excellent forecasting accuracy and the generalization is improved by using the pruning strategy. Applying this approach on the typical chaotic time-series : Mackey-Glass series, the results show that the precision and generalization of the algorithm with the new pruning strategy are improved. The algorithm can be applied without more prior knowledge, so it is suitable for practice.
出处
《华北电力大学学报(自然科学版)》
CAS
北大核心
2008年第2期88-93,共6页
Journal of North China Electric Power University:Natural Science Edition
基金
华北电力大学重大预研基金资助项目(20041306)
关键词
混沌时间序列
熵
自组织
模糊神经网络
chaotic time series
entropy
self-organizing
fuzzy neural networks