Vehicle Sales Forecasting Under A Combinatorial Model

14th Five-Year Plan of China has been fully launched,while the next five years will be an important period for building new advantages in the digital economy.Due to the acceleration of the transformation of the entire manufacturing industry,the automobile industry,as a representative of the high technology,intelligence and industrial concentration,has become a pioneer in the execution of intelligent manufacturing.At present,China has formed the world’s largest vehicle manufacturing consumer market,and the vigorous development of the vehicle industry is of great significance to driving domestic demand and export development.Therefore,predicting the demand for automobile sales accurately will not only help enterprise managers to control the overall development and growth of the market from a macro perspective,but also help automobile manufacturers study the market conditions from a micro perspective,formulate production and marketing strategies,choose forecasting methods reasonably,and forecasting demand accurately.Although there are many models and methods for sales demand forecasting,most of them are aimed at specific problems and specific fields.For enterprises,it is urgent to build ahighly personalized adaptability model for predicting the future demand of products from different angles.Based on the learning and understanding of domestic and foreign scientific research achievements,aiming at the defects and deficiencies of the traditional single model method,combined with the actual business requirements,and taking the actual provincial and monthly sales data of an automobile manufacturer as the breakthrough point,combined with the macro market,based on the sales data of competing products,a novel combination model is proposed from the perspective of coarse and fine granularity(provinces/stores).Under the provincial forecasting,one of the most popular methods in the field of sales forecasting at present,Light GBM algorithm is used as the main model.It is partially revised by combining the competitive analysis Holt-Winters model and the qualitative analysis model.Under the storelevel forecast,based on the provincial-level forecast,the more mature time series Holt-Winters model is used as the main model.The qualitative analysis model is combined to modify the model based on business analysis.At the end,combined models at two granularity levels are completed respectively.To further verify the actual performance of the model method of this paper in a specific business scenario of,this paper selects the mean absolute percentage error(MAPE)and the accuracy index provided by the business department of a major automobile manufacturer to demonstrate the superiority of the combination model respectively,which eventually provides the algorithm basis for the following development of the sales forecasting system.