馬可夫鏈蒙地卡羅法於機車零備件需求預測之研究

A Study on Demand Forecasting of Motorcycle Spare Parts: The Application of MCMC Method

張瀚陽(Han-Yang Chang)

指導教授 ： 張淳智

國立臺中科技大學/商管學群/流通管理系碩士班/碩士(2010年)

摘要

需求預測在機車供應鏈體系中扮演不可或缺的角色，尤其對於維持售後維修服務上更為重要。多數企業在進行需求預測時，多半會參考過往經驗，輔以簡單時間序列方法。然而，對於備用零件的特性而言，因為伴隨許多不確定因素，難以尋得特定規律，導致時間序列方法的預測結果往往無法有效地反映實際需求情況，進而造成無謂的庫存積壓與浪費。針對此一議題，過去卻少有文獻探討機車產業零件之需求預測問題。有鑑於此，本研究以機車零件供應商的角度，探討機車零件之使用壽齡對需求量之影響，建構一需求預測模型，並經貝式統計推導後，以馬可夫鏈蒙地卡羅法進行模式之校估，盼能最小化總預測誤差。
透過國內某機車零件總經銷商提供之資料庫，本研究挑選出16項機車零件，進而探討本研究模型與時間序列方法(移動平均、指數平滑、Croston法與修正Croston法)於零件需求之預測情況。結果顯示，比起各類時間序列方法，本研究模型更能反應實際之需求波動，且具有較低的預測誤差；此外，本研究亦比較馬可夫鏈蒙地卡羅法與基因演算法在參數估計上之差異。兩類的估計結果雖然相似，然在參數估計的過程中，MCMC法毋須像基因演算法一樣，必須透過多種的設定方能找到最佳近似解，在估計的流程上是更為直接且單純的。

關鍵字

備用零件 ； 需求預測 ； 貝式推論 ； 馬可夫鏈蒙地卡羅

並列摘要

Demand forecasting plays an important role in motorcycle supply chain, especially for maintaining certain after-sale service level. While there is a need of forecasting, most of companies choose to use simple time series method, and refer to their past experiences. As to the characteristics of spare parts, however, due to lots of undetermined factors which affect the accuracy of demand predicting, it is hard to find a certain regulation, and could easily result in an inaccurate demand predicting. In view of the fact that past studies, when discuss this problem, rarely focus on motorcycle industry, the purpose of the study is, on the motorcycle spare parts supplier’s point of view, to construct a forecasting model that is expected to minimize total predicting errors. The model considers the affect of ages of different spare parts toward demand, and is run an estimation process through Markov Chain Monte Carlo method (MCMC) to have the best parameters. Before that, the model is derived by Bayes’ theorem.
We choose 16 spare parts from the database of chef dealer of the country for the purpose of evaluating and comparing the forecasting power of our model and time series models (Moving average, Exponential smoothing, Croston, and SY Croston (Syntetos (2001)). As a result, our model could better fit the real demand patterns and has lower predicting errors than time series models mentioned above. The study compares parameter estimation methods of both MCMC and Gene Algorithm as well. We found that MCMC is much straightforward and simple that we just need to change sampling numbers to run the estimation process, while Gene Algorithm has to use different settings to make sure if the estimated parameters are nearly the best.

並列關鍵字

Spare Parts ； Demand Forecasting ； Bayesian Inference ； Markov Chain

參考文獻

吳沛軒 (民97)。考量零件生命週期下之汽車售後零組件需求預測與備貨模式。國立台灣大學商學研究所碩士論文，未出版，臺北市。

關智峰 (民98)。維修零件最終訂購問題之需求預測模式研究。國立台中技術學院流通管理系碩士班碩士論文，未出版，臺中市。

Chen, F. L., Chen, Y. C., & Kuo, J. Y. (2009). Applying Moving Back-propagationNeural Network and Moving Fuzzy Neuron Network to Predict the Requirement of Critical Spare Parts. Expert System with Applications, In Press, Corrected Proof, Available online 27 November.

Croston, J.D. (1972). Forecasting and Stock Control for Intermittent Demands.Operational Research Quarterly 23(3), 289–303.

Gelfan, A. E., & Smith, A. F. M. (1990). Sampling-based Approaches to Calculating Marginal Densities. J. Am. Stat. Asso. 85(410), 398-409.