Research And Implementation Of Convenience Store Data Analysis System Based On LSTM Network

With the rapid advance of the Internet and big data technology,many enterprises are searching for new strategies of management and business.The chain convenience store has become one of representatives of “new” retail business,how to manage the huge mass of business data growing dramatically and how to dig up valuable information from the mass data need to be addressed immediately.Based on the business data of a chain convenience store in Hunan,this paper designs and implements the chain convenience store data analysis system from four aspects: purchase,distribution,inventory and sales.The system not only provides the display of the business data,but also provides users with the statisti cal and analytical functions of data.Among them,for the massive sales data of chain convenience stores,this data analysis system designs and implements the sales forecast function of convenience stores.This function relies on the sales forecast model o f DAM_LSTM network based on multi-level attention mechanism proposed in this paper.This model improves the accuracy of forecasting by focusing on the time-cycle characteristics of sales data and the impact of non-forecasting data on forecasting data.Specifically,the attention mechanism is used to obtain the important influence of non-predictive columns on predictive columns in multivariate time series data sets.Inspired by the attention mechanism,an intermediate state LSTM network of encoder is constru cted to capture the information of the changing rules of time series data.Then the attention mechanism is used to learn the changing rules of predictive sequences over time.The LSTM network with intermediate state of decoder is built to capture the infor mation of temporal data variation.Finally,two attention models are integrated into LSTM-based network to construct a temporal prediction model based on DAM_LSTM network.Using the sales data of mineral water in the convenience store as experimental data set,this paper compares the proposed model with the traditional LSTM network time series prediction model and the LSTM(AM_LSTM)network time series prediction model based on simple attention mechanism.The experimental results show that the proposed model is superior to the other two models in terms of the accuracy of sales forecast and the attention paid to the characteristics of the data set.Based on the above-mentioned DAM_LSTM time series prediction model,this data analysis system realizes the function of category-oriented commodity sales prediction,and also achieves the function of statistics and analysis of diversified business data,enhances the management ability of chain convenience stores' business data,and improves the level of chain convenience stores' information management.