铁路路基冻胀的自适应定量预测模型  

Adaptive and quantitative forecast model of railroad subgrades' frost-heaving index

作　　者：吴湘华[1] 乐天晗 陈峰[1] 吴永军[1] 

机构地区：[1]中南大学交通运输工程学院,湖南长沙410075

出　　处：《铁道科学与工程学报》2017年第6期1154-1162,共9页Journal of Railway Science and Engineering

基　　金：铁道部重点资助项目(2012 G009-B);铁路总公司科技研究开发计划课题(2014G001-E)

摘　　要：针对严寒地区路基冻胀的定量预测,提出一种优化灰色与神经网络组合模型。采用设置时距权值矩阵、微分方程背景值优化和模型初值优化的方式,对传统的非等时距GM(1,1)预测模型进行优化,并对初始预测残差采用BP神经网络进行修正。选取哈大客运专线某区段2013-12~2014-01路基冻胀数据,利用该模型对其进行拟合与预测,所建立的冻胀预测模型精度值达到0.984,后验差比达到0.108 6,平均预测误差值1.46%,与现有GM(1,1)和BP网络模型相比,预测结果精度明显提高,实现了对路基冻胀较高精度的定量预测。According to the quantitative forecast of the railway’s frost-heaving data of cold region, the paperproposed a combined model of optimization gray and neural network. By adopting a time interval weight matrixand differential equation’s background value and model’s initial value, the traditional non-equal time intervalGM（1,1） forecast model was optimized. The combined forecast model of optimized grey and neural network byadopting BP neural network was established to rectify the initial forecast of residual errors. The model in thispaper was applied in matching and forecasting the frost-heaving data of subgrades which collected fromHarbin-Dalian Passenger Railway during the time period from December 2013 to January 2014. From the results,it is found that the posterior error ratio is only 0.108 6 and the average prediction error value is 1.46%, as themodel accuracy value as high as 0.984. It is more accuracy than the existing GM（1,1） and BP models, which helpsto realize the high-accuracy quantitative forecast of the frost-heaving.Key words： railway engineering; railroad subgrades’ frost-heaving index; forecast; optimal gray model; BPnetwork; combined forecast model

关 键 词：铁道工程 路基冻胀 预测 灰色优化 BP网络 组合预测 

分 类 号：U216[交通运输工程—道路与铁道工程]