基于选择性集成ARMA组合模型的零售业销量预测

Retail Sales Combination Forecasting Model Based on Selective Ensembled ARMA

摘要:

准确预测商品销量的走向对零售企业具有重要意义,构建自回归移动平均模型(ARMA模型,Auto-Regressive and Moving Average Model)对零售商品时序销量数据进行预测分析。传统ARMA模型无法准确描述商品销量中同时存在的非平稳非线性特征。论文分别采用支持向量回归(SVR,Support Vector Regression)方法和极限学习机(ELM,Extreme Learning Machine)方法,对时序模型中非线性误差进行预测并进行误差补偿,提高了商品销量的预测精度。提出了遗传优化的选择性集成定阶方法,用以简化ARMA模型的复杂定阶过程,降低了对数据平稳性程度要求。论文收集了某电商平台商品销量数据,对ARMA、选择性集成ARMA、ARMA-SVR、ARMA-ELM四种预测模型的性能进行了对比分析,结果表明,选择性集成ARMA模型预测精度在平稳和非平稳时序数据下分别提高23.58%和41.28%。组合模型相比仅采用线性平稳时序模型的预测结果更符合实际,其中,ARMA-SVR模型在小样本、非平稳时序下预测精度比ARMA-ELM模型高出约三分之一。

Abstract:

Abstract: It is important for retail enterprises to forecast the sales volume as accurately as possible.By building auto-regressive moving average model (ARMA) can forecast and analyze the retail sales volume. The traditional ARMA model can not describe the unstable and nonlinear characteristics which existing in the sales volume. The Support Vector Regression(SVR) method and Extreme Learning Machine(ELM) method are used to predict and compensate the nonlinear errors of ARMA time series model to enhance the accuracy of prediction. And the paper proposed a novel selective ensemble method to determine the parameter of ARMA.This method based on a genetic optimization algorithm and simplified the process of ARMA. We gathered some data to analyze and compare the performance of ARMA、selective ensembled ARMA、ARMA-SVR and ARMA-ELM. The experimental results show that the accuracy of selective ensembled ARMA model are increased by 23.58% and 41.28% in both stable and unstable series. The combination forecasting model is more in line with the actual data than time series forecasting model. Aslo, ARMA-SVR model’s accuracy of prediction is about one-third higher than ARMA-ELM .