基于二次分解和模型选择策略的港口集装箱吞吐量组合预测

A Combination Forecast Method of Port Container Throughput Based on Secondary Decomposition and Model Selection Strategy

准确预测港口集装箱吞吐量对于政府部门规划港口建设,港口和航运企业合理调配资源具有重要意义。已往研究往往采用单一分解方法来处理序列中的复杂特征,存在数据特征提取不完全以及预测模型选择比较盲目的问题,极大地影响了组合模型的预测效果。为此,本文引入二次分解和基于数据特征的模型选择策略,通过建立组合预测框架对港口集装箱吞吐量进行预测。首先,根据原始序列的整体特征选择一种分解方法对其进行初步分解,得到若干分量。然后,分析各分量的平稳性、季节性及复杂性等数据特征,据此选择合适的计量经济学模型进行预测或采用完全自适应噪声集合经验模态分解(CEEMDAN)方法对分量进行二次分解。接着,引入长程相关性特征,根据二次分解后子序列的平稳性、复杂性、长程相关性等再选择合适的预测模型。最后,将所有分量的预测结果集成从而得到最终的预测值。以月度预测为例,本文选取上海港和天津港集装箱吞吐量数据作为样本开展实证研究。实证结果表明,本文所提出的组合预测框架与基准模型相比具有更高的预测精度,是一种比较有前景的港口集装箱吞吐量预测工具,可以为相关政府部门、港口及航运企业提供决策参考。

Accurate prediction of port container throughput is of great significance to government planning of port construction and rea⁃sonable allocation of resources by ports and shipping enterprises.Previous studies often use a single decomposition method to deal with the complex features in the container throughput series,which results in incomplete data feature extraction and blind selection of predic⁃tion models,and greatly affects the prediction effect of the combined model.Therefore,this paper introduces the strategies of secondary decomposition and model selection based on data characteristics analysis to predict the port container throughput by establishing a combi⁃nation forecast framework.Firstly,according to the overall characteristics of the original series,a decomposition method is selected to de⁃compose it preliminarily,and several components are obtained.Then,the data characteristics of each component,such as stationarity,seasonality and complexity,are analyzed,and an appropriate econometric model is selected for prediction or the complete ensemble em⁃pirical mode decomposition with adaptive noise(CEEMDAN)method is used to decompose the component.Next,the characteristic of long⁃range correlation is introduced,and the appropriate model for prediction is selected according to the stationarity,complexity,and long⁃range correlation of the sub⁃sequences after the secondary decomposition.Finally,the predicted values of all components are inte⁃grated to obtain the final prediction results.Taking monthly forecast as an example,the empirical study uses the container throughput da⁃ta of Shanghai Port and Tianjin Port as samples.The empirical results show that the combination forecast framework proposed in this pa⁃per has higher prediction accuracy than the benchmark models,and is a promising tool for port container throughput forecasting,which can provide decision⁃making reference for relevant government departments,ports,and shipping enterprises.