In this research, we study flexible flow line scheduling problems with unrelated parallel machines and with a bottleneck stage. The measures of performances are to minimize makespan, to minimize the number of tardy jobs, and to minimize total tardiness, considered respectively. Several bottleneck-based heuristics are developed to solve these scheduling problems. A bottleneck-driven multiple insertion heuristic (BDMIH) is proposed to solve problems with makespan as the objective. The essential idea of BDMIH is that we think the scheduling of jobs at the bottleneck stage may affect the performance of a heuristic for scheduling jobs in all stages. Therefore, in this heuristic we let jobs entering the sequence at the first stage be driven by their sequence entering at the bottleneck stage. Given an FFL problem with a bottleneck stage, this heuristic first identifies the bottleneck stage, then generates an initial sequence of jobs by a variant of Johnson’s rule (SPT-LPT rule), and finally applies a multiple insertion heuristic to find the best schedule. Another heuristic, a bottleneck-based due-date decision heuristic (BBDDDH), is developed to solve problems with the number of tardy jobs as the objective. The heuristic consists of three steps: (1) Identifying the bottleneck stage, (2) Scheduling jobs at the bottleneck stage and the upstream stages ahead of the bottleneck stage, and (3) Using dispatching rules to schedule jobs at the downstream stages behind the bottleneck stage. A new approach is developed to find the arrival times of the jobs at the bottleneck stage, and two decision rules are developed to schedule jobs at bottleneck stage. This new approach neatly overcomes the difficulty of determining feasible arrival times of jobs at bottleneck stage. The last bottleneck-based heuristic, a bottleneck-driven adaptable multiple insertion heuristic (BDAMIH), is constructed to solve problems with total tardiness as the objective. The main idea of BDAMIH is combined with the ideas of BDMIH and BBDDDH. The main difference between BDAMIH and BDMIH is that BDMIH generates an initial sequence of jobs before performing the insertion heuristic; however, BDAMIH is adaptable to select a job within the process of the insertion heuristic. To evaluate the performance of the proposed heuristics, several well-known dispatching rules and heuristics are investigated for comparison purposes and computational experiments are performed on randomly generated test problems. Computational results show that the proposed heuristics significantly outperform all well-known dispatching rules or heuristics. Also, an analysis of the experimental factors is performed, and several interesting insights of the proposed heuristics are discovered.

本研究開發一個以區域最佳解為基礎的群體式 (population-based) 啟發式演算法(簡稱HLBS)，來求解流程式排程（flow shop）之最大流程時間的最小化問題。其中，HLBS會先建置一個跟隨模型來導引搜尋機制，然後，運用過濾策略來預防重複搜尋相同解空間而陷入區域最佳解的困境；但搜尋仍有可能會陷入區域最佳解，這時，HLBS則會啟動跳脫策略來協助跳出區域最佳解，以進入新的區域之搜尋；為驗證HLBS演算法的績效，本研究利用著名的Taillard 測試題庫來進行評估，除證明跟隨模型、過濾策略和跳脫策略的效用外，也提出實驗結果證明HLBS較其他知名群體式啟發式演算法(如基因演算法、蟻群演算法以及粒子群最佳化演算法)之效能為優。 / This research proposes population-based metaheuristic based on the local best solution (HLBS) for the permutation flow shop scheduling problem (PFSP-makespan). The proposed metaheuristic operates through three mechanisms: (i) it introduces a new method to produce a trace-model for guiding the search, (ii) it applies a new filter strategy to filter the solution regions that have been reviewed and guides the search to new solution regions in order to keep the search from trapping into local optima, and (iii) it initiates a new jump strategy to help the search escape if the search does become trapped at a local optimum. Computational experiments on the well-known Taillard's benchmark data sets will be performed to evaluate the effects of the trace-model generating rule, the filter strategy, and the jump strategy on the performance of HLBS, and to compare the performance of HLBS with all the promising population-based metaheuristics related to Genetic Algorithms (GA), Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO).