Prophet-VAR组合优化模型在高值卷烟销量预测中的应用  

Application of Prophet-VAR combined optimization model in predicting the sales of high-priced cigarettes

作　　者：康静 姚春玲 KANG Jing;YAO Chunling(China Tobacco Business Logistics Co.,Ltd,Beijing 100055,China;Beijing Zhongyan Information Technology Co.,Ltd,Beijing100055,China)

机构地区：[1]中烟商务物流有限责任公司技术部,北京100055 [2]北京中烟信息技术有限公司业务三部,北京100055

出　　处：《中国烟草学报》2023年第1期127-134,共8页Acta Tabacaria Sinica

摘　　要：针对高值卷烟销量时间序列的非平稳性、趋势性、周期性和节假日性等特点,同时考虑到高值卷烟销量受行业政策的影响,结合Prophet模型、VAR模型和单箱结构因子法,构建了Prophet-VAR组合优化模型。选用2011—2021年全国高值卷烟销量数据分析检验Prophet-VAR组合模型,结果表明,单独Prophet和单独VAR的预测精度分别是87.97%和84.30%,Prophet-VAR组合模型的预测误差值约为4%,精度接近96%,比单一模型的预测效果提高了约10个百分点。考虑单箱结构等行业政策因素的影响,运用单箱结构因子法对预测结果进行优化,使得预测精度达到了98.75%。因此,优化后的组合模型较单一的模型能更好地表现高值卷烟销量时间序列的变化趋势,给出更好的预测结果。The time series of high-priced cigarette sales is characterized by non-stationarity, tendency, periodicity and holiday. To this end,a Prophet-VAR combined optimization model based on Prophet, VAR and the single-box structure factor was constructed to improve the prediction accuracy of high-priced cigarette sales. The feaisliby of the Prophet-VAR combination model was verifed based on the sales data of high-priced cigarettes nationwide collected from 2011 to 2021. The results show that the prediction accuracy of Prophet alone and VAR alone are 87.97% and 84.30%, respectively. The prediction error of the Prophet-VAR combined model is about 4%, and the accuracy is close to 96%, which is at least 10 percentage points higher than that of a single model. Considering the influence of industry polices such as the single-box structure, the single-box structure factor is used to improve the prediction by multiplication, which results in an accuracy of 98.75%. Therefore, the combined optimization model can better represent the changing trend of the time series of high-priced cigarette sales than the two single models and can obtain better forecasts.

关 键 词：高值卷烟销量 时间序列 向量自回归 VAR PROPHET 

分 类 号：F224[经济管理—国民经济] F426.8