Research On Forecast Model Of Refined Sugar Sales Based On Multidimensional Features Abstract

In the 21 st century,with the continuous development of Internet technology and the continuous improvement of the level of informatization,in order to adapt to the increasingly fierce market competition,the information management model has been gradually applied to the sugar industry.The traditional sugar industry management model is gradually moving toward Informationization,automation management model transformation.However,in the process of transformation,the sugar industry has not fully utilized the historical sales data accumulated in the information system for many years,resulting in waste of data,resulting in inaccurate planning or unreasonable decision-making when the production plan of the finished sugar is formulated.The finished sugar produced has frequent product hoarding or shortage,which has brought huge economic losses to the company.Therefore,a more accurate forecast of the sugar sales of sugar companies can not only provide relevant decision support for enterprises,but also provide relevant technical support for the long-term development of enterprises.The main research contents of this paper have the following two points.1.The feature combination optimization algorithm GBDT-FO based on gradient lifting decision tree model is proposed to extract and optimize the multi-dimensional features of the finished sugar sales data.In this paper,the monthly sales data of a sugar company from 2016 to 2018 is used as the raw data for constructing the forecasting model.To verify the accuracy of the forecasting model,the monthly sales data of 2016 and 2017 are used as training data,monthly sales data of 2018 is used as verification data,and the sliding window method is used to increase the amount of data,and the characteristics of the originaldata are enlarged to construct the original data set.Then,the features in the original data are analyzed and extracted to obtain the initial multi-dimensional feature set,and the construction features are adopted.The combination method is used to improve the dimension of the feature space.The experimental comparison shows that the model performance of the feature combination is improved compared with that before the combination;then the GBDT-FO algorithm proposed in this paper is used to optimize the feature combination,which further enhances the dimension of the feature space.The comparative analysis of the experiment shows that the generalization ability of the model after feature combination optimization is significantly enhanced.Finally,the feature with high correlation with the prediction result is selected from the features of the combined optimization as the model input of the subsequent refined sugar sales forecasting model.2.A model of finished sugar sales forecast based on optimized combination single forecasting model is constructed.In this paper,the four basic sales forecasting models based on machine learning are optimized,and the original finished sugar sales forecasting model is constructed by means of mean processing.The problem of weak learning ability of single sales forecasting model is solved.The contribution degree of each basic optimization model to the forecasting target is measured,and the respective contribution coefficients are learned.Finally,the optimized finished sugar sales forecasting model is constructed.Through the experimental comparison and analysis,the original finished sugar sales forecasting model is better than the accuracy.The single basic optimization model,combined with the contribution coefficient,optimizes the refined sugar sales forecasting model to be better than the original finished sugar sales forecasting model.After verification,the finished sugar sales forecast model predicted by the finished sugar sales forecast model proposed in this paper is used.The data is basically in line with the real data of the monthly sales of finished sugar in 2018,and the monthly sales forecast dataof finished sugar in 2019 is given.