Research On Forecasting Models And Methods Of Rarely Used Spare Parts' Demand

Spare parts management plays an important role in industry equipment management. Spare parts management aims to cut down occupied capital and related cost so as to improve reliability, maintainability and economy of equipments, and is closely related with the manufacturing schedule and overall profit. Accurate forecast on spare parts demand is crucial to optimize spare parts management.Rarely used spare parts demand with limited history demand data samples is difficult to forecast with traditional statistic forecast methods, due to its appearance at random with many time periods having no demand. Based on the analysis of rarely used spare parts' data characteristics and the current methods which is used to forecast rarely used spare parts demands, the thesis introduces Syntetos' categorization scheme on demand patterns to classify rarely used spare parts demand into following patterns:erratic demand, internmittent demand and lumpy demand. Then, this dissertation designs new forecast methods for each demand pattern as follows:Firstly, to improve the forecasting accuracy of erratic demand, an ensemble empirical mode decomposition (EEMD) based hybrid modeling framework is proposed. This approach is under a "decomposition-and-ensemble" principal to decompose the original erratic demand series into several independent smooth subseries including a small number of intrinsic mode functions (IMFs) and a residue by EEMD technique. Then support vector machine regression (SVR) based forecasting methods are used to model each of the subseries so as to achieve more accurate forecast respectively. Finally, the forecasts of all subseries are aggregated to formulate an ensemble forecast for the original erratic demand series. This approach is called "EEMD-SVM" forecasting method.Secondly, a Modulation Forecasting method was designed to forecast the time when the intermittent demand occurs. An intermittent demand series could be decomposed into two subseries:one is demand size subseries, the other is a "0-1" subseries with "1" denote a demand occurs. A carry wave was designed to transform the "0-1" series into a continuous and smooth series. This series is forecasted by EEMD-SVM method. Finally a detector was designed to detect the "0" and "1" from the forecast. An example is raised to verify the rightness and the effectiveness of the method. The result shows that this method can forecast the demand occurs time at an ideal accuracy. Thirdly, a combination forecast method based on EEMD-SVM method and Modulation Frecasting method was designed to forecast intermittent demand and lumpy demand. Some artificial data and real spare parts demand data were used to compare this combination forecast method with other common used methods such as moving average, single exponential smoothing, Croston method, SVR etc. The result shows that the accuracy of this method is fairly better than other methods.Lastly, a forecasting support system assembling multiple methods is designed for rarely used spare parts demand. It is called as "Non-normal Demand Forecasting Support System" (NDFSS). NDFSS can automatically classify the rarely used spare parts' demand, choose the fit forecasting method and optimize the parameters of chosen method intelligently.