Food Supply Chain Demand Forecasting Based On Machine Learning

With the further development of informatization and industrialization in the world,traditional independent enterprises have been gradually replaced by different supply chains,making modern business competition gradually evolve from the enterprise level to the supply chain level.The entire supply chain starts from raw material suppliers,through manufacturers,wholesalers,distributors and various retailers,and finally extends to the customer group.During the operation of supply chain,there is the flow of funds,commodities and information.The ultimate purpose of the establishment of the supply chain is to meet the basic needs of customers and to achieve the goal of the lowest overall cost and the greatest overall benefit of the enterprise in the supply chain.How to compress inventory costs to the greatest extent without reducing customer favorability has become an inevitable problem in the development of the supply chain.The common form of supply chain demand forecasting is mainly based on the historical data of target commodity sales,and on this basis,it is reasonably extrapolated.The forecasting accuracy of the model is mainly affected by the experience of forecasters.Or use the statistical analysis model to analyze and predict the time series model of sales records,which requires a lot of manpower and material resources in data collection,model update and analysis.This paper uses machine learning theory,combined with the current supply chain demand forecasting development status,to propose a series of improvement measures on the supply chain demand forecasting system,including input variable selection,model parameter adjustment,and combination forecasting weight optimization algorithm.The preliminary screening of input variables is achieved by establishing the correlation between variables,ensuring that the recommended model can extract the most critical modeling information from the massive raw data;the optimization scheme for model parameters is based on different models themselves Attribute selection,using optimization algorithms to quickly and accurately adjust model parameters to reduce model deviations due to human factors;the model combination is to determine the combination weight by combining the fitting performance of different models on the data set.Better increase the model's tolerance to extreme data and improve the model's robustness.The case analysis part takes G company's food supply chain terminal demand data as an example to conduct an empirical analysis study on the improved demand forecasting analysis framework.The experimental results show that the extreme gradient lifting model has the best effect when forecasted by a single model,and can be predicted by a combination of ways to further improve the accuracy of the model and provide some help for future research.