运筹与管理2025，Vol.34Issue(2)：38-43,6.DOI:10.12005/orms.2025.0040
新能源汽车销量预测的分解—聚类—集成方法研究
Sales Forecasting of New Energy Vehicles with a Decomposition-cluster-ensemble Method
摘要
Abstract
The monthly sales data of new energy vehicles have the phenomenon of multi-data characteristics such as nonlinear and seasonal aliasing,and the use of a classical single model for forecasting has the disadvantage of low prediction accuracy.To improve the accuracy of monthly sales forecast of new energy vehicles,based on the modeling idea of"decomposition-ensemble",on the basis of making full use of the advantages of each single model,and following the principle of"divide and conquer",a comprehensive prediction model of"decomposition-clustering-ensemble"is constructed to achieve high-precision prediction of monthly sales of new energy vehicles. Firstly,the ensemble empirical mode decomposition(EEMD)model is applied to decompose the time-series data of monthly sales volume of new energy vehicles.This approach effectively handles the nonlinear and non-stationary data characteristics of the data series,and effectively suppresses the occurrence of modal aliasing phenomenon.Then,to improve the efficiency of prediction modeling and reduce the accumulation of errors,sam-ple entropy and K-means method are used to cluster the obtained decomposition components,and three types of components are obtained:high frequency sequence,medium frequency sequence and low frequency sequence.Based on the advantages of GM(1,1)model,which is suitable for the prediction of exponential law data series,the low frequency class component series is predicted.The autoregressive integrated moving average(ARIMA)model can transform complex non-stationary sequences into stationary sequences for modeling,and use it to predict intermediate frequency component sequences.The long short-term memory(LSTM)network model selects and processes the data through the three gates inside the control,which is suitable for more complex high-frequency data series prediction modeling,and uses it to predict the high-frequency component series.Finally,the linear weighting method is used to combine the forecast results of each component,and the forecast results of monthly sales of new energy vehicles are obtained. The monthly sales volume of new energy vehicles from January 2012 to May 2022 published by the China Association of Automobile Manufacturers is used as the data set to verify the"EEMD-K-LSTM/ARIMA/GM(1,1)"comprehensive forecasting model proposed in this study.The results show that compared with the traditional single model and the"decomposition-ensemble"model,the"decomposition-clustering-ensemble"comprehen-sive forecasting model achieves a good forecasting effect,and the MAPE value of the one-step forward and three-step forward forecasting of the monthly sales volume of new energy vehicles is 8.75％and 10.62％,respectively.Using the"EEMD-K-LSTM/ARIMA/GM(1,1)"comprehensive forecasting model,the sales data of new energy vehicles in China from January 2012 to October 2022 are modeled.The predicted sales for November 2022 to January 2023 are 800,000 vehicles,830,000 vehicles,and 520,000 vehicles,respectively,consistent with the overall trend in 2020 and 2021. It should be noted that in reality,there are multiple factors that influence the monthly sales of new energy vehicles,including national policies,seasonal factors,economic conditions,and so on.To achieve long-term trend forecasting,the next step should consider incorporating various influencing factors into the model and conducting more comprehensive predictions and discussions through methods such as scenario analysis.关键词
新能源汽车/销量预测/EEMD分解/K-means聚类/分解—集成Key words
new energy vehicles/sales forecast/EEMD decomposition/K-means clustering/decomposition-ensemble分类
管理科学引用本文复制引用
王方,赵桉坤,卜皓玥,余乐安..新能源汽车销量预测的分解—聚类—集成方法研究[J].运筹与管理,2025,34(2):38-43,6.基金项目
国家自然科学基金资助项目(72001165) （72001165）
滨州魏桥国科高等技术研究院"产学合作协同育人"项目(BINTECH-KJZX-20220831-67) （BINTECH-KJZX-20220831-67）
陕西省创新能力支撑计划资助项目(2022KJXX-112) （2022KJXX-112）
西安市科技计划项目软科学研究重点项目(23RKYJ0006) （23RKYJ0006）
