Research On The Model Algorithm Of Sales Forecast Based On ConvLSTM Network In The New Retail Format

With the continuous rise of the new retail industry,the sales environment has become more complex.The key to how to take advantage of its digital,omni-channel,and flexible supply chain to scientifically guide enterprise management and operations lies in the prediction of user needs.For the new retail industry,there are many reasons that affect the sales of goods,and the variety of goods is const antly increasing.The mutual influence of goods makes the traditional demand forecasting method not suitable for forecasting the demand for goods in the context of new retail.Therefore,there is an urgent need to study a suitable demand forecasting method to accurately forecast demand based on the characteristics of the new retail industry and increase the company's inventory management and control capabilities.This article takes real product sales data of the new retail industry as the research object,analyzes the existing demand forecasting methods,and combines the Word2 Vec model to correlate the clustering of the commodities according to the sales characteristics of the new retail industry,and explores the related commodities based on the Conv LSTM network The feasibility of the sales forecast model.First,it analyzes the current research status at home and abroad,and summarizes it from two aspects of demand forecasting and deep learning.It is found that the traditional demand forecasting method has some drawbacks,it needs to construct a large number of feature engineering,the model reuse is poor,and the association relationship of the commodity cannot be learned.In this paper,based on the deep learning Conv LSTM network,a sales prediction model for associated commodity is proposed.Then,the sales forecasting model of the entire related product is constructed in two parts.The first part constructs a product association clustering model based on the Word2 Vec model,clusters the related products,uses the clustering results for the input of the Conv LSTM network model,and introduces prior knowledge of product associations in the input.The second part,based on the Conv LSTM network,builds a related product sales prediction model.The Conv LSTM net work can use convolution operations to extract data space characteristics and LSTM to obtain long-term time-dependent characteristics of data to predict product sales.Finally,combined with the real sales data of Q company,the experimental application of the sales prediction model based on Conv LSTM network was carried out and compared with the prediction models such as LGBM and LSTM.Throughout the process,we conducted data exploration of sales data,detailed the entire process from data preprocessing to model training,and analyzed the experimental results.The results show that the Conv LSTM network can effectively extract the association characteristics between products,and carry out information transmission on the time series of product sales,making the model more robust and improving the model prediction effect.