在工業工程領域，生產排程一直受到顯著的關注。當企業的業務部門決定接收哪些
訂單時，必須同時考量本地工廠與外包工廠的產能，以制定一個全面的計劃。然而，目前的文獻中很少同時探討訂單選擇排程與外包選項的研究，這可能低估或高估工廠的總產能進而造成公司巨大損失。為了彌補這些差距，本研究提出了一個模擬優化模型，透過考量本地工廠的排程限制與外包工廠的返還時間，進行訂單選擇。該模型確定了需在特定工序中外包的訂單，並為本地工廠提供生產排程。此外，為了增強模擬系統的真實性與穩定性，將訂單生產時間以及外包返還時間設為隨機變數。本研究基於數位孿生的框架下，建立一個能夠符合實時性 (Real-time Capability)、動態調整能力 (Dynamic Adaptability)、多層次決策支持 (Multi-level Decision Support)的演算法架構，其結合機器學習、傳統啟發式演算法以及分層優化結構，並在解決目標問題方面有良好的表現。本研究的貢獻如下： 1. 建立一個模擬系統 ，可以模擬根據不同訂單選擇、外包決策、與本地排程計算工期與利潤 。 2. 提出一個結合機器學習與啟發式算法的演算法架構，能為該問題在較短時間內找到優良的柏拉圖前沿解 。 3. 模擬不同外包返還時間變異與外包成本對系統利潤造成之影響，並給出管理決策洞見。

In the field of industrial engineering, production scheduling has consistently received significant attention. When a company's business unit decides which orders to accept, it must simultaneously consider the capacities of both local and outsourced factories to develop a comprehensive plan. However, existing literature rarely addresses the integration of order selection, scheduling, and outsourcing options, which may lead to underestimation or overestimation of factory capacity, potentially resulting in substantial losses for the company.
To address these gaps, this study proposes a simulation-optimization model that incorporates local factory scheduling constraints and the return times of outsourced factories for order selection. The model determines which orders need to be outsourced at specific production stages and provides a production schedule for the local factory. Furthermore, to enhance the realism and stability of the simulation system, order processing times and outsourcing return times are treated as stochastic variables. This research developed within the framework of a digital twin, establishes an algorithmic architecture that adheres to key characteristics such as real-time capability, dynamic adaptability, and multi-level decision support. The architecture combines machine learning, traditional heuristic algorithms, and hierarchical optimization structures, demonstrating strong performance in addressing the target problem.
The contributions of this study are as follows: 1. Development of a Simulation System: A system capable of simulating makespan and profit based on different combinations of order selection, outsourcing decisions, and local scheduling. 2. Proposal of an Algorithmic Framework: A framework integrating machine learning and heuristic algorithms that efficiently identifies high-quality Pareto front solutions for the problem within a shorter timeframe. 3. Impact Analysis of Outsourcing Return Time Variability: Simulation of the effects of variability in outsourcing return times on system profitability, providing valuable decision-making insights.