本文爲德國班貝克大學的碩士論文，共86頁。
本論文是與西門子Healthineers合作撰寫的。目標是預測X射線系統供應鏈的未來銷售數字。更好地瞭解未來的銷售數字有助於更好地分配資源。該生產線由五個不同的系統組成，所有系統的銷售數字都是以單個量進行預測的。提供的歷史數據涵蓋15年時間，包括180個數據點。這些數據點是與銷售有關的每月普查數據，將傳統的統計時間序列建模技術與新興的機器學習方法進行了比較。採用的時間序列建模技術是指數平滑和ARIMA。機器學習技術包括建模前饋神經網絡和隨機森林。所有方法的性能均用平均絕對百分率誤差進行測量。所有實現方法的最佳性能來自於擴展的ARIMA模型（ARIMAX），相對誤差達到9.55%。此外，本文還包括了一個供應鏈管理軟件工具的實現。
This thesis was written in cooperation withSiemens Healthineers. The goal was to predict future sales figures of X-raysystems for the supply chain management. Better knowledge about future salesnumbers enables better allocation of resources. The product line consists offive different systems. Sales figures are predicted for all systems as a singlequantity and also individually. The historic data supplied covered 15 years andconsisted of 180 data points. These data points are a monthly census of salesrelated figures. Traditional time series modeling techniques from statisticswere compared with newly emerging machine learning approaches. The establishedtime series modeling techniques were exponential smoothing and ARIMA. Themachine learning techniques consisted of modeling feedforward neural networksand random forests. The performance of all methods was measured with the meanabsolute percentage error. The best performance of all implemented methodsresulted from an extended ARIMA model (ARIMAX) with a relative error of 9.55%.Moreover, the thesis included implementing a software tool for the supply chainmanagement to make forecasts in practice.
- 引言
- 理論背景與以前工作回顧
- 研究方法
- 評估方法
- 結論
附錄A 實驗
完整資料領取請加QQ羣免費下載：