优化GM(1,1)与SVM组合模型的路基冻胀预测应用  

The forecast application of subgrade's frost-heaving index based on optimized GM(1,1) and SVM combination model

作　　者：乐天晗 吴永军[2] 吴湘华[2] 陈峰[2] 

机构地区：[1]广州地铁设计研究院有限公司,广东广州510010 [2]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11期2345-2351,共7页Journal of Railway Science and Engineering

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

摘　　要：基于非等时距GM(1,1)优化预测模型,采用支持向量机进行预测残差修正,建立一种组合预测算法,并运用该算法对铁路路基冻胀进行定量预测。对经典非等时距GM(1,1)模型背景值和初值的计算方法进行优化,同时设置时距权值矩阵,对不同时间测量所得数据赋予不同权重。在初始预测后,对残差值采用支持向量机进行非线性修正,得到最终预测值。选取哈大客专某区段实际测量路基冻胀数据,对算法实用效果进行检验。所建立预测模型平均预测误差值为2.039%,最大预测误差5.911%,后验证差比值0.005,各项指标均优于单一灰色模型与文献[6]中建立的组合预测模型,实现了对铁路路基冻胀的较高精度定量预测。The paper proposed a forecasting algorithm combining the optimized non-equal interval GM（1,1）model and the adopted support vector machine to rectify the initial forecast of residual errors. This algorithm wasthen used to forecast the railway frost-heaving data quantitatively. It optimized the calculation methods oftraditional non-equal time interval GM（1,1） forecast model’s differential equation’s background value and model’sinitial value. The time interval weight matrix was set, and different weights on the data measured at different timeswere assigned. The adopted support vector machine was applied to rectify the initial forecast of residual errorsto get the final forecast. The model in this paper was applied in matching and forecasting the frost-heaving data ofsubgrades that are collected from Harbin-Dalian Passenger Railway. The model’s posterior error ratio is only0.005. The average prediction error value is 2.039% and the maximum prediction error value is 5.911%. It is moreaccurate than the current GM（1,1） models and the combined forecasting algorithm proposed in the literature. Thus,the highly accurate quantitative forecast of the frost-heaving was achieved.

关 键 词：铁道工程 路基冻胀 支持向量机 灰色模型 组合预测 残差修正 

分 类 号：U216[交通运输工程—道路与铁道工程]