面向深度学习的商品销售额预测研究  

Research on commodity sales forecast oriented on deep learning

作　　者：唐甜甜 周伟[1] TANG Tiantian;ZHOU Wei(School of Mathematics and Physics,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学数理学院,兰州730070

出　　处：《重庆理工大学学报（自然科学）》2022年第7期310-316,共7页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(61863022);中国博士后科学基金项目(2017M623276)。

摘　　要：由于销售数据的非平稳性等问题,深度学习模型很难预测未来值,为此,提出了一种新的转换平稳特征的方法。为了量化模型预测的准确性,引入了结合提取平稳特征能力的CNN和分析时间序列能力的LSTM,构建串联CNN-LSTM的预测模型,与XGBoost、支持向量机(SVM)、LSTM和CNN模型在测试集上通过拟合曲线并计算指标以作出评判。决策出影响销售额的关键因素,为企业改进促销方式,进而提高收益,具有现实意义。Due to the non-stationarity of sales data and other issues,the DL model is difficult to predict future values.In this paper,a new method of transforming the stationary characteristics is proposed,which enables the DL model to be applied effectively.In order to quantify the prediction accuracy of the model,a tandem CNN-LSTM prediction model is built by combining the ability of extracting stationary features with the ability of analyzing time series with the ability of CNN and LSTM.The prediction model is fitted with XGBoost,SVM,LSTM and CNN models on the test set and calculated the indicators to make a judgment.It is of practical significance to determine the key factors that affect the sales,and then improve the promotion method and increase the income for enterprises.

关 键 词：CNN-LSTM 预测分析 深度学习 平稳特征 

分 类 号：F222[经济管理—国民经济]