考慮隨機能耗限制及未知偏好之智慧多準則生產規劃

Abstract

本研究係針對未知決策者偏好並考慮隨機能耗限制下之智慧多準則生產規劃進行探討，以協助決策者有效找到滿足其多準則偏好的決策。由於未考量隨機能耗限制下的定性生產排程本身就是NP-hard問題，同時隨機問題在實務上並沒有一個很有效的求解方法；因此，本研究提出一項三階段的解決方案，包含1) 隨機能耗機率模型建構, 2) 未知決策者偏好下之互動式多準則最佳化演算法, 3) 考慮機台隨機能耗機率限制下的細部排程演算法。 本研究提出之未知決策者偏好下之互動式多準則最佳化演算法(iMOPUP)，主要是在偏好函數形態不明下，協助決策者準確並有效地在多個相互衝突的決策目標下找到最適解。在本研究中這個增強的決策支持方法，被應用來制定最合適的生產派工策略，在未知決策者偏好下，所提出的互動式演算法透過與決策者進行少量的互動，即使在決策者無法有效辨識方案間的差異時，也能夠有效地協助決策者找出平衡相互衝突的製造準則下之最佳策略。 本研究提出之考慮生產機台能耗的細部排程啟發式演算法（SFDS），目的是要同時提高生產效率和能源效率。這個強化的啟發式排程方法，能夠在滿足工廠和電廠簽訂的最大額定功率下，不超約並有效地制定最佳的細部生產排程。透過隨機能耗機率模型，來模式化工件與機台間相依的能耗關係並捕捉機器能耗的隨機特徵，將這些能耗特徵應用於細部生產排程最佳化的過程中，從而避免超約。 在考慮能耗限制並應用生產現場收集的數據驗證下，iMOPUP及SFDS兩項演算法表現都優於其他演算法。數值模擬及分析的結果也顯示本研究所提出的演算法不論在求解品質及計算時間上都有優越的表現。從每次與決策者互動中，iMOPUP都會嘗試學習決策者的決策偏好並提供對應的方案。數值分析結果顯示，iMOPUP能夠只透過與決策者約十次的互動，就找到在線性或二次偏好函數下的近似最佳解。在二次和線性偏好函數下，iMOPUP與最佳解之間的平均差距分別僅小於2％和1％。SFDS在最高能耗限制下，平均排程總製造時間保持在最佳解的0.6％以內。在規模大的問題中，SFDS的排程總製造時間與最佳解之間的平均差距僅為0.02％；此外，當問題的規模增加時，SFDS所需的計算時間依然保持穩定。

This research investigates multi-criteria production planning problems under unknown user preference functions, to help decision-makers (DMs) find solutions that meet their multi-criteria preferences. As the deterministic version of the dispatching or scheduling problem without stochastic energy consumption constraints is NP-hard, no practical-sized stochastic problem can likely be solved efficiently. Thus, the proposed method is designed to decompose into 1) stochastic energy modeling, 2) interactive method for multi-criteria optimization problems (MOP) under Unknown preference functions, and 3) detailing of scheduling problems of machines under probabilistic energy constraints. An interactive method for MOP under Unknown Preference (iMOPUP) is proposed to help DMs find solutions accurately and efficiently under multiple conflicting objectives with the unidentified form of the preference function. This enhanced decision support method was adopted to determine the most preferred dispatching policy that balances the conflicting manufacturing criteria through a small number of interactions with unknown DMs’ preference, provided that DMs are indistinguishable when selecting the preferred solution among alternatives. An energy-aware Shop Floor Detailed Scheduling heuristics (SFDS) is also proposed to enhance both production and energy efficiency. This enhanced heuristic was investigated through the production scheduling problems under maximum power consumption constraints. Probabilistic models are built to model dispatching-dependent and capture the stochastic characteristics of machine energy consumption. The energy consumption characteristics are then used to optimize the scheduling and thereby avoid exceeding the energy usage constraints. The proposed methods, iMOPUP and SFDS under power consumption constraint, were validated through real manufacturing data and results reveal the proposed methods outperform other methods. Numerical results also suggest the superiority of proposed methods in terms of solution quality and computational time. The proposed iMOPUP can address the continuous criteria space without requiring the feasible nondominated solution set in advance. In every interaction, iMOPUP attempts to learn the DMs’ preferences and provides the corresponding alternatives. The numerical study reveals that iMOPUP finds near-optimal solutions under linear or quadratic utility functions through less than a dozen interactions with DMs. The average gap between the iMOPUP solutions and true optimal is merely less than 2% and 1% under quadratic and linear utility functions, respectively. The proposed SFDS, while constraining the probability of violating peak energy consumption constraints, the makespan of the scheduling solution remains within 0.6% of the optimal solution on average. In large-scale problems, while the total energy consumption is optimal, the gap between SFDS and the optimal solution of the makespan is merely 0.02% on average. Moreover, the required computational time remains stable while the problem size increases.