Research And Realization Of Household Appliances Demand Forecast Based On Multi-model Fusion

Demand forecasting refers to the assessment made by an enterprise according to the expected demand for the future market of a product.Demand forecasting is the basis of supply chain plans for all industries,and supply chain decisions are made before the actual occurrence of demand.The planning department of an enterprise needs to use historical demand information and experience to predict future demand,and adjust the factory capacity at the front of the supply chain and inventory at the back end according to these forecasts results.With the development of data science,enterprises have the ability to collect and store data,and the traditional experience-based demand forecasting can not completely eliminate the bullwhip effect in the supply chain.For this reason,the demand for forecasting based on data mining has emerged and has been recognized by more and more enterprises.It is of great significance to use data mining technology to build demand forecasting models.However,there are still exist many shortcomings in the current demand forecast of the company’s supply chain.On the one hand,the data produced by the supply chain has the characteristics of non-standard and complex business logic.On the other hand,the algorithm principle of demand forecasting is different.If we only use a single model for forecasting,it is likely to cause the loss of useful information.This paper analyzes the production and sales data of kitchen appliances of a household appliance enterprise,and constructs a household appliance demand forecasting model based on multi-model fusion.The main works of this paper are as follows:(1)In order to solve the trouble of complex and numerous data in the supply chain,data cleaning operations are carried out before modeling work in this paper.On the one hand,this paper combs all the demand-related factors from production to retail in the supply chain,and analyzes and preprocesses the target data through Data Visualization.On the other hand,this paper also subdivides the data into three main feature groups: time,product attributes and target volume trends for feature extraction through combining business logic which lays the data foundation for subsequent predictive modeling work.(2)In order to delete redundant features and reduce the difficulty of model training,this paper uses a variety of feature screening methods,and compares the prediction effects on external evaluators.Experiments show that the all of the three feature screening methods can improve the prediction effect to a certain extent,and the method based on SHAP(SHapley Additive ex Planation)has the best effect.(3)This paper uses the filtered features to construct three demand prediction models based on LSTM(Long Short-Term Memory),RF(Random Forest),and Light GBM(Light Gradient Boosting Machine)respectively.In the model training process,Bayesian Optimization is introduced to optimize the model parameters to reduce the disturbance of the parameters to the model.Experiments show that the accuracy of single-model forecasting has little difference and has a certain degree of generalization,but the performance is different when the demand fluctuates greatly.(4)In this paper,learners(LSTM,RF,Light GBM)with certain predictive effects are used as the base model for model fusion,and the model fusion method is Stacking.In order to further improve the accuracy of prediction model,and follow the principle of Accurate but Diversified,this paper selects LSTM,RF,Light GBM as the base model,and uses MLR(Multiple Linear Regression)as the meta-model,builds a model of demand forecast based on the Stacking fusion model.The experimental results show that the R-squared of this the Stacking fusion model can reach 0.7629,and the forecast error RMSE is 0.1303,which is significantly better than the other three base models.Multi-model fusion makes full use of the advantages of each model,and improves the prediction effect and generalization performance of the model.(5)This paper take the above-mentioned feature engineering,parameter tuning,and multi-model fusion as the core,designs and implements a prototype system for home appliance demand forecasting,and provides a feasible solution for the application of demand forecasting for home appliance companies.In conclusion,this paper proposes some technical improvement measures in view of the inherent shortcomings of the demand forecast of home appliance companies.This paper firstly sorts out the impact factors of demand forecasting through data analysis,then improves the accuracy of a single model through feature selection and parameter tuning,and finally uses the method of multi-model fusion to further improve the forecast accuracy.It shows that under the guidance of the current methodology,in the supply chain demand forecasting business scenario,excellent forecasting results and rich management enlightenment can be obtained,which proves that the demand forecasting framework based on the supply chain proposed in this paper has greater application value.