Research On Short-term Demand Forecasting Method Of E-commerce Commodities Based On Multi-layer Hybrid Deep Neural Network

Owing to the accurate demand forecasting can not only help enterprises formulate reasonable replenishment plans and inventory decisions to reduce inventory costs,but also improve the efficiency of supply chain operation and meet consumer experience,the Short-term demand forecasting of commodities is a key task in e-commerce supply chain management,However,the short-term demand forecasting of e-commerce commodities has become a difficult and key problem in the field due to the dynamic,intermittent and complex factors of commodity trading in e-commerce environment.In this background,this paper studies the research on commodity demand forecasting by domestic and foreign scholars,and systematically analyzes the problems to be solved in demand forecasting model and content of e-commerce commodity firstly.Secondly,the paper defines and explains the concepts and characteristics which are related to the short-term demand of ecommerce commodities.Meanwhile,the paper studies the related content and forecasting method theory of short-term demand forecasting of e-commerce commodities,which provides the basic foundation for subsequent forecasting modeling.In addition,through the comparative study of forecasting methods,as well as combing with the relevant characteristics and influencing factors of short-term demand of e-commerce commodities,a multi-layer hybrid deep neural network(AR-MDN)model based on previous research results is established,which can simultaneously learn and extract feature sets,simulate time series trends and demand probability distribution.Furthermore,exploratory data analysis method and feature engineering theory are used to pre-process the collected historical transaction data of e-commerce commodities,and the original feature clusters needed for the prediction model is constructed.Lastly,the prediction effect of the multi-layer hybrid deep neural network model is verified and the paper also compares the prediction results with ARIMA model and MLP-LSTM model.By forecasting the short-term demand of e-commerce commodities with the established model,the results show that the forecast value of short-term demand of e-commerce commodities based on AR-MDN model fits well with the real value and the forecast error is stable,which indicates that the model has good accuracy.Since the constructed feature set takes into account some derived features,the paper compares the effects of derivative features on the prediction results of AR-MDN model.It is found that the derived features play a certain role in improving the accuracy of the model.Secondly,compared with the prediction results of the two comparison models,the AR-MDN model has lower root mean square error(RMSE)and mean absolute percentage error(MAPE)in short-term demand forecast of e-commerce commodities and the prediction results of the AR-MDN model is superior to the other two models in regional warehouse forecasting,which indicates that AR-MDN model has better accuracy and robustness in short-term demand forecasting of e-commerce commodities.