改进BP算法在公路工程主材价格预测中的应用研究  

Research on Application of Improved BP Algorithms in Price Prediction for Highway Engineering Main Material

作　　者：徐家兵[1] 祁志国[1] 杭文[1] 何杰[1] 李旭宏[1] 

机构地区：[1]东南大学交通学院,江苏南京210096

出　　处：《公路交通科技》2008年第4期62-66,共5页Journal of Highway and Transportation Research and Development

基　　金：安徽省2006年交通科技进步计划资助项目(2006-34)

摘　　要：分析了传统BP算法的不足,利用相关分析法筛选出公路工程主材价格的主要影响因素;在确定BP神经网络结构及选取训练函数的基础上,建立了基于改进BP神经网络算法的公路工程主材价格预测模型,并结合合肥市石屑价格预测的实例,利用建立的预测模型,采用BP传统算法及附加动量法、自适应学习速率法、两者相结合法等3种改进算法分别预测了合肥市2个季度的石屑价格,并将预测结果进行对比,分析了不同BP算法预测结果之间的差异。结果表明,使用改进的BP神经网络算法进行公路工程主材价格预测,可以将预测误差控制在6%以内,并减少95%左右的训练步数。同时采用自适应学习速率和附加动量改进BP网络的方法相对最有效。The prices of main materials for highway engineering are complicated variables which are affected by many factors. Application of improved BP algorithms in price prediction of highway engineering main materials was discussed. The specific methods are as follows： First, the shortcomings of traditional BP algorithm were pointed out and then its improved algorithms were put forward. Then, the main influencing factors of highway engineering main materials were screened out with correlation analysis method. On the basis of determining the structure of BP neural network and selecting training functions, a price forecasting model based on improved BP algorithms of highway engineering main materials was established. Finally, combining with the case of forecasting the prices of stone chips in Hefei, traditional and three improved BP algorithms were used and the forecasting results were analyzed. The results indicate that forecasting by applying improved BP neural network is feasible and effective. The forecasting errors can be reduced to less than 6% and training steps can be reduced by 95% percent using self-adaptive study speed algorithm and additional momentum algorithm simultaneity.

关 键 词：道路工程 工程主材 BP神经网络 自适应学习速率 附加动量法 价格预测 

分 类 号：U414[交通运输工程—道路与铁道工程] F407.9[经济管理—产业经济]