Research On Support Vector Machines Based Forecasting Method For Rarely Used Spare Parts Demand

Spare parts management is one of the important tasks of equipment management, which ties up with the operating performance and economical efficiency of the enterprise. The basic principle of spare parts inventory management is not only to fulfill the requirement of equipment maintenance and overhaul, but also to occupy reasonable capital. Accurate forecast of spare parts demand is crucial for spare parts inventory control and optimization.Rarely used spare parts with intermittent demand which appears at random, with many time periods having no demand, have very few demand data samples. So it is very difficult to forecast the demand of spare parts of this type. Forecasting methods based on traditional statistics can not solve this problem properly. According to this, the thesis does some research on support vector machines(SVMs)based forecasting method for rarely used spare parts demand.First of all, the intermittent demand characteristics of rarely used spare parts are described, and then the basic ideas and drawbacks of the intermittent demand forecasting methods in common use such as exponential smoothing, Croston method and Bootstrap method etc. are analyzed.Secondly, the forecasting method of support vector machines which based on the statistical learning theory is introduced. And the linear and nonlinear support vector regression (SVR) algorithms are also explained.Thirdly, two kinds of SVR based forecasting methods for rarely used spare parts demand are depicted, they are SVR forecasting method based on time series and SVR forecasting method based on impact factors. The main steps and frame of these methods and the estimation approaches of forecast results are discussed. And then an example is given to verify the validity of the SVR based forecasting methods for rarely used spare parts demand.Lastly, a SVR based forecasting support system (FSS) for rarely used spare parts demand is designed.