模糊聚类神经网络的非对称学习算法

通过仿真和分析表明模糊聚类神经网络原有学习算法FCNN的局限性，如初值敏感性、吸引域不灵活和稳定点不合理等；指出造成上述局限的原因主要在于算法的对称性和权值向量的修正缺乏协同。为此，通过在网络模型中引入层内反馈、在算法中引入加速项，消除了算法的对称性并使权值向量的修正具有一定的协同性；通过改进算法结构，消除了小尺度振荡现象并使算法的稳定点趋于合理。计算机仿真结果表明改进后的非对称学习算法AFC可以有效克服原有算法的不足并具有较高的收敛速度。

AN ASYMMETRIC ROBUST LEARNING ALGORITHM OF FUZZY CLUSTERING NEURAL NETWORKS

Demonstrated by simulations and analysis are weaknesses of the current learning algorithm of fuzzy clustering neural networks FCNN, such as initial condition sensitiveness, micro scale vibration, inflexible attracting basin and unreasonable convergence points. Furthermore, it is pointed out that the causes of the above weakness mainly lie in the symmetry of the algorithm and the lack of cooperation among the modifications of net weights. In order to eliminate the above weaknesses, the inner layer feedback is introduced, the structure of the algorithm is improved, and a robust algorithm AFC is proposed. The result of simulations verifies the proposal properly.