基于GM(1,1)-NGO-SVM模型的浙江省铁路货运量预测研究

Research on Railway Freight Volume Prediction in Zhejiang Province Based on GM(1,1)-NGO-SVM Model

摘要: 铁路货运量作为货运物流领域的核心衡量指标，其精准预测对于优化货运资源配置、指导铁路部门运营决策具有重要价值。单一预测模型在预测实践中，往往存在信息全面性不足的问题。为了提高铁路货运量预测的精准性，本文创新性提出了一种基于GM(1,1)-NGO-SVM的组合预测模型，旨在克服单一预测模型的不足。选取浙江省1982年~2017年的铁路货运量数据作为训练样本，分别构建GM(1,1)模型与SVM模型，通过GM(1,1)模型捕捉数据序列的线性特征，通过SVM模型捕捉数据序列的非线性特征。为了进一步优化SVM模型的性能，引入北方苍鹰算法(NGO)对SVM模型的超参数进行智能寻优。最后，根据两个模型的平均相对百分比误差的倒数，采用加权求和的方式，构成了GM(1,1)-NGO-SVM组合预测模型。为了验证该模型的预测效果，本文选取了浙江省2018年~2022年的货运量进行预测，并与传统的SVM、NGO-SVM、GM(1,1)模型进行比较。实验结果表明：本文提出的GM(1,1)-NGO-SVM模型对浙江省铁路货运量的预测结果更加精确，同时可以更好预估浙江省铁路货运量未来的发展趋势，为浙江省铁路部门提供运营决策上的帮助。

Abstract: As a core measurement indicator in the field of freight logistics, the accurate prediction of railway freight volume is of great value for optimizing freight resource allocation and guiding operational decisions of railway departments. Single prediction models often suffer from insufficient information comprehensiveness in practical forecasting. To improve the accuracy of railway freight volume prediction, this paper innovatively proposes a combined prediction model based on GM(1,1)-NGO-SVM, aiming to overcome the limitations of single prediction models. Taking the railway freight volume data of Zhejiang Province from 1982 to 2017 as training samples, GM(1,1) and SVM models are constructed separately. The GM(1,1) model is used to capture the linear characteristics of the data series, while the SVM model captures the nonlinear characteristics. To further optimize the performance of the SVM model, the Northern Goshawk Optimization (NGO) algorithm is introduced to intelligently optimize the hyperparameters of the SVM model. Finally, based on the reciprocal of the average relative percentage error of the two models, a weighted summation approach is adopted to form the GM(1,1)-NGO-SVM combined prediction model. To validate the prediction performance of this model, this paper selects the freight volume data of Zhejiang Province from 2018 to 2022 for prediction and compares it with traditional SVM, NGO-SVM, and GM(1,1) models. The experimental results show that the proposed GM(1,1)-NGO-SVM model provides more accurate predictions of railway freight volume in Zhejiang Province and can better estimate the future development trend, thereby assisting railway departments in Zhejiang Province in operational decision-making.