基于FOA优化混合核LSSVM的铁路货运量预测

发布时间：2018-11-06 08:13

【摘要】：单一核最小二乘支持向量机(LSSVM)在铁路货运量预测中难以准确描述货运量的复杂变化特征,限制了预测精度的提高。针对该问题,提出一种基于果蝇算法(FOA)优化混合核LSSVM的预测方法。以多项式核与径向基核组合的混合核函数作为LSSVM核函数,构建铁路货运量的混合核LSSVM预测模型,同时利用FOA全局寻优能力强、计算速度快等优点优化选择混合核LSSVM参数。以我国铁路货运量为例进行方法验证。结果表明,所提方法的RMSE、MAE、MAPE和THEIL值分别为8 433.0、6 670.8、0.018 0和0.011 7,均小于其他模型,FOA算法搜索混合核LSSVM参数的时间为40.294 8 s,分别比GA和PSO算法减少了2.620 8 s和20.701 6s,适合于铁路货运量的短期预测。
[Abstract]:Single kernel least squares support vector machine (LSSVM) is difficult to accurately describe the complex variation characteristics of freight volume in railway freight volume prediction, which limits the improvement of prediction accuracy. To solve this problem, a hybrid kernel LSSVM prediction method based on Drosophila algorithm (FOA) is proposed. Using the hybrid kernel function of polynomial kernel and radial basis kernel as the LSSVM kernel function, the hybrid kernel LSSVM prediction model of railway freight volume is constructed, and the hybrid kernel LSSVM parameters are optimized by using the advantages of strong global optimization ability of FOA and fast calculation speed. Take our country railway freight volume as an example to carry on the method verification. The results show that the RMSE,MAE,MAPE and THEIL values of the proposed method are 8 433.0 / 6 670.8 / 0.018 0 and 0.011 7, respectively, which are smaller than those of other models, and the time of searching the LSSVM parameters of hybrid kernel by FOA algorithm is 40.294 8 s. Compared with the GA and PSO algorithms, it is 2.620 8 s and 20.701 6 s less respectively, which is suitable for the short-term prediction of railway freight volume.
【作者单位】：石家庄铁道大学经济管理学院;曼彻斯特城市大学商学院;北京大学光华管理学院;
【基金】：国家自然科学基金青年基金资助项目(61503261) 河北省软科学研究计划项目(15456106D) 河北省高等学校青年拔尖人才计划项目(BJ2014097) 河北省社会科学发展研究课题项目(2015020206) 河北省交通运输厅2015年科技计划项目(Y-2010024) 国家留学基金委(CSC)公派留学地方合作项目(201608130165) 河北省高校人文社会科学重点研究基地石家庄铁道大学工程建设管理研究中心资助项目河北省软科学工程建设管理研究基地资助项目河北省重点学科(管理科学与工程资助项目)
【分类号】：U294.13;TP18

【相似文献】