【摘要】 提出了一种基于聚类的神经网络算法 ,可以很好解决大样本情况引起的网络结构复杂、收敛性和泛化能力差等神经网络的固有问题 .算法采用聚类算法为分类器 ,进行模式空间分解 ,以分类后的模式子空间为各样本集合 ,用神经网络集学习 ,最后根据重力模型计算检测样本对各样本子集的隶属度 ,整合各子空间的输出结果 .通过实验对比表明该算法精度较高 ,容错性好更多还原

【Abstract】 A new prediction algorithm, called neural network bas ed on clustering was presented. In this algorithm, sample space was divided into s ub-space. A set of neural network was used to study every sub-space for predic tion separately. The subjections to every sub-space of the test sample were cal culated according to gravity model, and all prediction results were integrated i nto one. This new method can be used to slove some internal problems such as com plex net structure, weak constringency and pan-ability. The contrast experiment shows the method has high precision and good performance.更多还原