基于FGM-SVR组合模型的港口吞吐量预测  

Port Throughput Prediction Based on FGM-SVR Combined Model

作　　者：邓萍 刘淑龙 DENG Ping;LIU Shulong(School of Traffic&Transportation,Chongqing Jiaotong University,Chongqing 400074,China;School of Economics&Management,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]重庆交通大学交通运输学院,重庆400074 [2]重庆交通大学经济与管理学院,重庆400074

出　　处：《重庆交通大学学报（自然科学版）》2023年第8期132-138,共7页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：国家社会科学基金项目(16CJL054)。

摘　　要：为提高港口吞吐量的预测精度,建立基于分数阶累加GM(1,1)预测模型FGM(1,1)和支持向量回归(support vector regression, SVR)的组合预测模型进行港口吞吐量的预测。首先,分别运用FGM(1,1)模型和SVR模型对吞吐量进行预测;然后,针对传统组合模型赋权不能兼顾各单项模型在各时点预测能力强弱的问题,提出基于诱导有序加权平均(induced ordered weighted averaging, IOWA)算子的赋权方法进行组合预测;最后,以重庆港2005—2020年港口货物吞吐量为数据样本进行实例验证,分别使用FGM(1,1)模型、SVR模型和赋权后的组合模型进行港口吞吐量预测,并比较3种模型的预测精度;最后,分别使用这3种模型对2021—2024年港口吞吐量进行了预测。研究结果表明:基于IOWA组合预测模型的均方根误差、平均绝对误差和平均绝对百分比误差均明显小于各单一预测模型。该组合模型可以为港口吞吐量预测提供一种新的方法。In order to improve the prediction accuracy of port throughput,a combined prediction model based on fractional order cumulative forecast GM(1,1)prediction model FGM(1,1)and support vector regression(SVR)was established to predict the port throughput.Firstly,FGM(1,1)model and SVR model were applied to predict the port throughput respectively.Then,aiming at the problem that the traditional combined model weighting could not take into account the predictive power of each individual model at each time point,a weighting method based on of induced ordered weighted averaging(IOWA)operator was proposed for combination prediction.Finally,taking the cargo throughput of Chongqing Port from 2005 to 2020 as the data sample for example verification,the FGM(1,1)model,SVR model and weighted combined model were used for port throughput prediction,respectively.The prediction accuracy of the three models were compared and these three models were used to respectively predict port throughput from 2021 to 2024.The research results show that the root mean square error,mean absolute error and mean absolute percentage error of the combined model based on IOWA are significantly smaller than those of each single prediction model.The proposed combined model could provide a new method for port throughput prediction.

关 键 词：交通运输工程 港口吞吐量 组合预测 分数阶累加 支持向量回归 诱导有序加权平均算子 

分 类 号：U691.71[交通运输工程—港口、海岸及近海工程]