| Demand forecasting of petrol products in gas stations is crucial to the planning of initiative distribution of petrol products,especially to the stability of product supply in petroleum companies.The key to the implementation of initiative distribution is to forecast the oil demand of each gas station.Therefore,it is necessary to establish a methodology that can accurately predict the demand for the next distribution cycle.It will provide decision-making support for the scientific formulation of the oil company’s initiative distribution plan and ensure the stable supply of petrol products in gas stations.With the development of Internet of Things(IoT),gas stations have been deployed with the dynamic monitoring systems of the oiling machine and liquid level instruments.Hence distribution centers can easily access the data of Point of Sales and the stock levels of gas stations.Relying on the advantage of integration of data from all customers,distribution centers can predict future demands,and then accurately estimate the time when the stock level reaches the minimum safe stock level and the peak demand time windows to arrange a reasonable delivery time to prevent stock-outs.Due to the sales data are characterized by high overlap and strong noise and the daily sales curve of petrol products are characterized by non-Gaussian and high variations,so they bring considerable challenges to the demand forecasting of petrol products.In terms of this problem,the main work of this research is as follows:(1)Analysis of demand forecast based on data flow in gas stations.This paper analyzes the characteristics of demand forecasting based on data stream for petrol products in gas stations and the difficulties brought by forecasting the demand.According to the characteristics,the forecasting model of petrol products in gas stations is proposed.(2)The demand forecasting of petrol products based on clustering and nonparametric regression.First the scheme demand forecasting for the gas station is given.The proposed scheme uses the K-means algorithm to divide the sales data into multiple disjointed clusters,evaluates the clustering result of the daily sales curve of a product by seven validity indices and determines the optimal number of clustering.Then,the forecasting model using non-parametric regression is built for each cluster.Next,because the demand patterns are affected by many factors,this paper uses the decision tree to describe the relationship between the demand pattern of petrol products and the related factors,and realizes pattern match for the next distribution cycle in the future.Finally,the future demand curve can be depicted by calling the corresponding prediction model.(3)Case study and applied research.The forecast method is validated by using 120 days sales data of a gas station in Dalian City,which proves that the proposed method is an effective demand forecasting method of petrol products.This method provides a reasonable time window for each gas station in the next distribution and improves the scientificity and rationality of the distribution plan.In this study,the method of integrating the clustering validity indexes is proposed to determine the optimal clusters.Aiming at the characteristics of non-Gaussian and high variability of sales curve after clustering,a data-driven non-parametric regression model is proposed,which improves the accuracy of demand forecasting for the gas station.This method provides a reasonable distribution time window for each gas station in the next distribution,which improves the scientificity of the distribution plan and provides the decision support for the oil companies to realize initiative distribution of petrol products. |
