基于RN和GN的两种RBF神经网络的工程造价预测模型  

Two Engineering Cost Prediction Models of RBF Neural Network Based on RN and GN

作　　者：叶青[1] 陈菲[1] 刘婧[1] 

机构地区：[1]华侨大学土木工程学院,福建厦门361021

出　　处：《数学的实践与认识》2016年第7期25-30,共6页Mathematics in Practice and Theory

基　　金：中央高校基本科研业务费专项资金资助项目(JB-ZR1162);华侨大学高层次人才科研启动项目(12BS131);泉州市科技计划项目(2013Z31);留金发[2013]3050;福建省自然科学基金计划项目(2013J01194);泉州市科技计划项目(2014Z116)

摘　　要：建立了调用NEWRB函数的正规化网络RN和基于K-means聚类的广义网络GN的两种RBF‘神经网络的工程造价预测模型,以55个厦门市工程造价案例进行实证分析.结果表明:当调用NEWRB函数构建RBF模型时,其性能主要取决于分布宽度,而基于K-means聚类的RBF神经网络主要取决于重叠系数和隐含层节点数;基于广义网络GN的RBF神经网络模型的训练效果较差,但学习速度更快、预测精度更高.Two engineering cost prediction models of RBF neural net, normal network RN calling the NEWRB function and the generalized network GN based on K-means clustering, were built in this paper, using 55 cases of engineering cost in Xiamen to do the empirical analysis. The result shows that when the NEWRB function was called to build RBF model, its performance mainly depends on the width of distribution, while the RBF network based on K-means clustering mainly depends on overlap coefficient and the number of nodes in hidden layer. The training effect of RBF neural network model based on the generalized network GN is poor, but the learning speed is faster and has a higher prediction accuracy.

关 键 词：工程造价 RBF神经网络 正规化网络RN 广义网络GN 

分 类 号：TU723.3[建筑科学—建筑技术科学] TP183[自动化与计算机技术—控制理论与控制工程]