本研究的貢獻在於提出一個更加貼合企業實際使用情況的新產品動態採購
問題決策方法。現存許多關於需求預測的方法或許手段各異，卻往往因新產品缺乏所需歷史資料的關係而無法實際應用，為此一些研究提出利用產品間相似性預測新產品需求的方法，這些方法使用與新產品相似的過去產品資料解決了新產品缺乏歷史資料的問題，爾後由 Ban et al. (2019)提出的 Residual tree 則在此基礎上使用相似產品的 Covariates 並整合情境樹與最佳化，從單純的新產品需求預測進一步到了動態採購策略的規劃，而本研究延此思路提出了使用 Neural gas(Martinetz &Schulten, 1991)這一 Covariate-free 的演算法學習新產品需求情境樹以使用數學規劃求得採購策略的方法。一項跨足各行業領域的研究調查指出，新產品帶來的利潤大約占企業總體利潤的百分之二十五 (Cooper &Edgett, 2012)，足見新產品對於一間企業的重要性，雖該研究對象主要為美國企業，卻也不失為一項有參考性的指標，因而如何選用適當的方法對新產品進行更好地採購決策規劃對企業來說是相當重要的議題。本研究以成本表現與運算時間兩項指標作為主要探討方法可行性與實務價值的依據，經模擬實驗分析，我們所提出的方法於前述兩項指標上的結果確實不遜於作為比較對象的 Residual tree 演算法，對於新產品的動態採購規劃提供一個更具實務價值的選項。

The contribution of this study is to propose a method for making decisions on dynamic procurement problems for new products that is more suitable for the actual use of companies. Existing methods for demand forecasting may be based on different approaches, but they are often not practically applicable due to the lack of historical
data required for new products. For this reason, some studies have proposed methods to forecast the demand for new products using the similarity between products, which
solves the problem of the lack of historical data by using data of similar products sold in the past. Later, the residual tree algorithm proposed by Ban et al. (2019) uses
covariates of similar products and integrates scenario tree with optimization to move from merely making demand forecasting to procurement decisions for new products.
Our study extends this idea by proposing the use of the neural gas algorithm(Martinetz&Schulten, 1991), which is covariate-free, for learning new product demand trees to
make procurement decisions using mathematical programming. A cross-industry study has shown that new products account for approximately 25% of a company's total profit(Cooper &Edgett, 2012), which shows the importance of new products to a company. Although the study is focused on U.S. companies, it still has some reference value. Therefore, it is essential for companies to choose an appropriate method to make better procurement decisionsfor new products. In this study, the feasibility and practical
value of the proposed method are based on two indicators, cost performance and computation time, and the results of the proposed method are comparable to those of the residual tree algorithm. Based on the simulations, the performance of our proposed method on the two indicators is indeed as good as the residual tree algorithm as a
comparator, providing a more practical option for the dynamic procurement problem of new products.

第一章 緒論 1
第二章 動態採購與隨機規劃 3
第一節 新產品採購最佳化問題 3
第二節 情境樹 4
第三節 Residual tree 演算法 7
第三章 情境樹學習 12
第一節 最佳化建構方法 12
第二節 Neural gas method 14
第四章 模擬分析 21
第一節 實驗設計 21
第二節 分析結果 23
第五章 結論 31
第六章 參考文獻 33

Baardman, L., Levin, I., Perakis, G., &Singhvi, D. (2018). Leveraging Comparables for New Product Sales Forecasting. Production and Operations Management, 27(12), 2340–2343.
Ban, G. Y., Gallien, J., &Mersereau, A. J. (2019). Dynamic procurement of new products with covariate information: The residual tree method. Manufacturing and Service Operations Management, 21(4), 798–815.
Calfa, B. A., Agarwal, A., Grossmann, I. E., &Wassick, J. M. (2014). Data-driven multi-stage scenario tree generation via statistical property and distribution matching. Computers and Chemical Engineering, 68, 7–23.
Cooper, R. G., &Edgett, S. J. (2012). Best Practices in the idea-to-launch process and its governance. Research Technology Management, 55(2), 43–54.
Defourny, B., Ernst, D., &Wehenkel, L. (2011). Multistage stochastic programming: A scenario tree based approach to planning under uncertainty. Decision Theory Models for Applications in Artificial Intelligence: Concepts and Solutions, 97–143.
Fattahi, M., &Govindan, K. (2020). Data-Driven Rolling Horizon Approach for Dynamic Design of Supply Chain Distribution Networks under Disruption and Demand Uncertainty. Decision Sciences, 0(0), 1–31.
Fei, X., Gülpınar, N., &Branke, J. (2019). Efficient solution selection for two-stage stochastic programs. European Journal of Operational Research, 277(3), 918–929.
Høyland, K., &Wallace, S. W. (2001). Generating Scenario Trees for Multistage Decision Problems. Management Science, 47(2), 205–336.
Hu, K., Acimovic, J., Erize, F., Thomas, D. J., &VanMieghem, J. A. (2019). Forecasting new product life cycle curves: Practical approach and empirical analysis. Manufacturing and Service Operations Management, 21(1), 66–85.
Kouwenberg, R. (2004). Scenario generation and stochastic programming models for asset liability management. Polyhedron, 23(17 SPEC.ISS.), 2659–2664.
Latorre, J. M., Cerisola, S., &Ramos, A. (2007). Clustering algorithms for scenario tree generation: Application to natural hydro inflows. European Journal of Operational Research, 181(3), 1339–1353.
Martinetz, T. M., Berkovich, S. G., &Schulten, K. J. (1993). “Neural-Gas” Network for Vector Quantization and its Application to Time-Series Prediction. IEEE Transactions on Neural Networks, 4(4), 558–569.
Martinetz, T. M., &Schulten, K. (1991). A “Neural-Gas” Network Learns Topologies. In Artificial Neural Networks (Vol. 1, pp. 397–402).
Ponomareva, K., Roman, D., &Date, P. (2015). An algorithm for moment-matching scenario generation with application to financial portfolio optimisation. European Journal of Operational Research, 240(3), 678–687.
Turner, S., &Galelli, S. (2016). Building a reduced scenario tree for multi-stage stochastic programming. https://cran.r-project.org/web/packages/scenario/vignettes/buildtree.html
Xu, B., Zhong, P. A., Zambon, R. C., Zhao, Y., &Yeh, W. W. G. (2015). Scenario tree reduction in stochastic programming with recourse for hydropower operations. Water Resources Research, 51(8), 6359–6380.