模糊对向传播神经网络的学习算法由输入层至竞争层的连接权向量和竞争层到输出层的连接权向量两部分的学习组成.对于前者,分别选用聚类法和梯度下降法,本文研究了模糊对向传播神经网络的两种学习算法,并且从理论上分析了这两种算法的性质.把算法应用于著名MackeyGlass混沌时间序列预测问题中,实验结果表明后一种算法的学习精度及泛化能力较前一种算法要好,但前者的学习速度要快 The learning algorithm of fuzzy counterpropagation neural network consists of two parts: the connection weight vector from input layer to competitive layer and the connection weight vector from competing layer to output layer.For the former, clustering method and gradient descent method are used respectively, Two kinds of learning algorithms of fuzzy counterpropagation neural network are proposed, and the properties of these two algorithms are theoretically analyzed. The algorithm is applied to the famous MackeyGlass chaotic time series prediction problem. The experimental results show that the latter algorithm’s learning accuracy and Generalization ability than the previous algorithm is better, but the former learning faster