Research And Application Of Retail Enterprise Sales Forecast Based On Stacking Model Fusion

Sales forecast in modern enterprises is the core link of goods supply.It is essential for the reasonable control of production costs,arranging the production plan of goods,and adjusting the inventory of goods.It is an important reference for marketing,logistics,production and other links.The subject research uses historical sales data of enterprises to provide data support for enterprise production and business decision-making,forecasts enterprise sales and integrates with existing enterprise information management platforms,and effectively utilizes data and related forecast results.The main research contents are as follows:1)Design and implementation of Stacking fusion model.According to the characteristics of Stacking integrated learning,RF(Random Forest),SVR(Support Vector Regression),ARIMA(Autoregressive Integrated Moving Average,differential integrated moving average autoregressive),LGBM(Light Gradient Boosting Machine),GRU(Gated Recurrent Unit,gated recurrent unit)five single models are primary learners,and CatBoost is a fusion model of secondary learners,which predicts the product sales of enterprise stores.The experimental results show that the final prediction result of the Stacking fusion model is smaller than the best single model CatBoost,the RMSE(root mean square error)is smaller,from 0.0194 to 0.0166;the MAE(mean absolute error)is smaller,from 0.0158 to 0.01650.0144;the value of R~2(coefficient of determination)is larger,from 0.8648 to 0.9005.2)Program design of sales forecast system.A database structure and application program are designed that use grid search to automatically adjust parameters,and then perform training and prediction based on the best parameters.Using the feature that the primary learner in the Stacking model can run in parallel,the program is deployed in a distributed manner,and the program is started regularly in the free time of the system to automatically complete the prediction task.It solves the disadvantage that the Stacking model needs to spend a lot of time to adjust the parameters after fusing multiple models.3)Enterprise information management platform integration.Combined with the current status of enterprise informatization,a distributed microservice architecture based on Spring Cloud Alibaba was built.Integrate the management platform with the ERP system,retail system and sales forecast system to obtain the sales data in the retail system on a daily basis and feed back the forecast results to enterprise users for viewing.In practical applications,the prediction results of the Stacking fusion model are generally consistent with the trend of the true value,and the accuracy rate is relatively stable.The integration with the enterprise information management platform not only solves the problem of data interoperability in the existing system,but also allows the enterprise to check the sales and sales status at any time.The forecast results provide convenience and data support for enterprises to make efficient and correct decisions.