结合优先度排序神经网络的样本点分块拟合研究  

Study on samples regression based on priority ordered neural network

作　　者：朱世交[1] 杨珺[1] 

机构地区：[1]上海电力学院电子与信息工程系,上海200090

出　　处：《计算机应用》2009年第3期862-864,共3页journal of Computer Applications

基　　金：上海市重点科技攻关计划资助项目(071605125)

摘　　要：样本点拟合是动态构造神经网络应用的重要研究领域。从输入输出映射的角度,依赖输出样本点优先关系,结合优先度排序神经网络把输入样本集合划分为不同子集,动态构造优先度排序神经网络对各个子样本集进行映射,对样本点进行分块并行神经网络构造,提高神经网络的训练速度。最后,通过对不同类型样本集进行测试,实验结果表明该算法能有效地减少拟合误差。Samples regression is an important issue in the field of Dynamic Constructive Neural Network(DCNN).From the view of the mapping values of neural network,input side and output side training samples were separated into different sub-sets according to the algorithm based on Priority Ordered Neural Network(PONN),which was a more similar model of human-being and widely used in different fields because of its priority neurons in neural network.Then,sub-sets were constructed into PONN using a parallel algorithm wh...

关 键 词：优先排序神经网络 拟合 拓扑空间 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]