基于支持向量回归的地铁牵引能耗预测  

Forecasting traction energy consumption of metro based on support vector regression

作　　者：陈垚[1] 毛保华[1] 柏赟[1] 冯瑜[1] 李竹君[1] 

机构地区：[1]北京交通大学城市复杂系统理论与技术教育部重点实验室,北京100044

出　　处：《系统工程理论与实践》2016年第8期2101-2107,共7页Systems Engineering-Theory & Practice

基　　金：国家重点基础研究发展计划项目(2012CB725406);国家自然科学基金(71571016;71131001)~~

摘　　要：预测地铁线路未来牵引能耗.有助于评价线路的牵引用能效率、节约能源.地铁牵引能耗影响因素众多且呈非线性关系.因此基于历史数据建立支持向量机回归模型对地铁牵引能耗进行预测.首先,将牵引能耗的影响因素分为供电系统、线路条件、列车属性、运营组织及环境因素五类,并选取线路可变影响因素作为模型输入;然后,利用遗传算法对模型参数进行寻优,适用度函数设计采用交叉验证方法:最后,基于模型最优参数对牵引能耗进行预测.案例结果表明,交叉验证方法有助于提高模型预测精度;支持向量机回归模型的预测精度高于BP(back-propagation)神经网络模型与多元线性回归模型.Forecasting traction energy consumption of metro contributes to evaluate the energy efficiency of lines and save traction energy. Traction energy consumption is nonlinearly affected by multiple factors. In this paper, a support vector regression （SVR） model based on the historical data was proposed for metro traction energy consumption prediction. First, the influencing factors on traction energy consumption were classified to power supply system, train characteristics, track profiles, operation scheme and meteorological factors. The variable factors of metro lines were chosen as input data. Thereafter, the genetic algorithm （GA） with cross validation was applied to optimize the parameters of the SVR model. Lastly, the SVR model with the optimal parameters was utilized to forecast the traction energy consumption of a metro line. The forecasting results indicate that cross validation improves prediction accuracy of the SVR model and that the SVR model achieves higher prediction accuracy than the back-propagation neural network （BPNN） model and the multiple linear regression （MLP） model.

关 键 词：地铁 牵引能耗 支持向量机 遗传算法 

分 类 号：U231.92[交通运输工程—道路与铁道工程] TK011[动力工程及工程热物理]