Research And Implementation Of Methods To Support Auto Parts Sales Forecast

With the increasing sales volume and quantity of automobiles,the requirement for auto parts in the process of automobile production and after-sales service has also increased sharply.The auto parts market in China will also usher in a new round of outbreaks.Faced with the huge market of auto parts,China's auto parts companies want to seize market share and gain profits in the fierce auto parts market competition environment In addition to improving the quality of auto parts continuously,they also need to greatly improve the capabilities of the collaborative management capabilities and reduce the management cost of auto parts.Although auto parts companies have deployed the coordination of supply chain,and used warehousing systems to improve the management level in recent years,the timely supply rate of auto parts and the inventory backlog are always the two major problems that troubled auto parts companies.In order to solve these problems,companies need to obtain useful information from a large amount of historical sales data to predict the market demand effectively for auto parts.For the demand forecast of auto parts,there are some problems that lead to low prediction accuracy.For example,the forecasting model is single,the feature selection of the features is not caused and so on.Therefore,this paper researches the auto parts sales data accumulated by companies in the past ten years of “ASP/SaaS-based manufacturing industry value chain collaboration platform”,which belongs to the national key research and development program “Distributed Resource Giant System and Resource Synergy Theory”(Projection No: 2017YFB1400301).On the basis of analyzing the problems and needs of auto parts sales forecast,the paper completes the design of platform-oriented auto parts sales forecasting.Among them,for the missing values,outliers,and irregular data formats in the sales data of auto parts,the data processing tools Pandas and Numpy are used to clean and convert the historical data of auto parts sales,which provides guarantee of data quality for the subsequent research work.Based on the analysis of the characteristics of auto parts,this paper studies the advantages and disadvantages of different feature selection methods,and proposes a two-stage feature selection method based on Filter and Wrapper mode.This method is used to select the features from data of auto parts sales in the platform.Aiming at the auto parts with short replacement period and large data scale,a prediction model based on LSTM(Long Short Term Memory)is proposed.A multi-model fusion prediction model based on machine learning is proposed for auto parts with long replacement period and small data scale.The algorithm is verified experimentally.