Study On Seasonal Forecasting Models And Applications For Short Time Series

In a demand-driven management mode,accurate demand forecasting has become a key to efficient operations of supply chain and enhancement of enterprise’s competitiveness.It is particularly important for fast moving consumer goods industry with seasonal characteristics.The accuracy of seasonal demand forecasting can affect the balance between supply and demand and the social harmony and stability to a great extent.However,food preferences of consumers are constantly changing as customer lifestyles change,and new products are frequently emerging as the enterprise’s capabilities in research and development are enhanced.Thus,the history of data used for product demand forecasting is very short. In terms of the time series data with short history and scarce information,the existing traditional forecasting model cannot directly be employed.Only the prediction technology for short time series data can solve the problems caused by the lack of information.Existing seasonal forecasting methods for short time series include :Individual Seasonal Indices method( ISI),Grouping Seasonal Indices method(GSI)and Shrinkage Seasonal Indices method(SSI).The GSI and SSI can improve the prediction precision of the ISI method to a certain extent.However,no studies have shown whether combining two methods can play a greater role or not.Therefore,in order to discover more accurate seasonal demand forecasting methods,a theoretical analysis on the feasibility of the combination are firstly carried out.Two novel demand forecasting models called JS-GSI and GSI-JS are explored and are compared with the existing methods theoretically.Then,a series of simulation experiments are designed to inve stigate the application rules of the models to the actual data,Finally,a case study based on beer sales data is carried out.The results show that the new models can improve the prediction accuracy to a great extent. It is described as follows:Chapter One introduces the research background,literature review,research content,technical route,research methods and research significance;In Chapter Two,the basic theoretical study of the forecasting model are carried out,which includes definition,fundamental principle and classification of forecasting.The attention is paid to the introduction of the three existing seasonal forecasting methods.On the basis of the existing models,two novel seasonal forecasting models are proposed in Chapter Three. The performance of all models are compared,and emphasis is laid on the examination of three factors which affect the performance of the models based on MSE as an evaluation criterion. In Chapter Four, a simulation experiment is designed.The effects of three main factors on the performance of the new models are analyzed. In Chapter Five,an empirical research based on the data from a large beer company in Wuhan is conducted,which further verified the results of the theoretical research.The best prediction model for the beer demand forecasting is selected. Chapter Six summarizes the research findings and the innovation points,and points out its future research direction.