Archive-Based Optimization Heuristics for Fuzzy Multi-Objective Unrelated Parallel Machine Scheduling Problems

• Main Authors: Ai-Lian Jhong, 鍾愛蓮
• Other Authors: Chiuh-Cheng Chyu
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/44057636397991246854

碩士 === 元智大學 === 工業工程與管理學系 === 97 === 　　This research studies unrelated parallel machine scheduling problems (UPMSP) with two objectives – minimizing average tardiness and the number of tardy jobs. To make the study better fit a real world application, job processing times are defined as triangular fuzzy numbers whereas due dates as trapezoidal fuzzy numbers. Three state-of-art algorithms, multi-objective simulated annealing (MOSA), Pareto archived evolution strategy (PAES), and archived multi-objective simulated annealing (AMOSA), are applied to solve the bi-objective UPMSP. Each algorithm consists of three decoding schemes, one of which is the commonly used decoding scheme - machine-group assignment with fixed weighted vectors (MGA_FW) on objectives, and the other two are novel schemes: (1) machine-group matching improvement with fixed weighted vectors (MGMI_FW); (2) machine-group matching improvement with random weighted vectors (MGMI_RW). 　　Benchmark instances with problem size 100 (jobs) x 5 (machines) and 100 x 10 using three levels for each of the two parameters, due-date tightness and due-date range, are generated according to a method in the literature. Our experimental results indicate that the decoding scheme MGMI_RW excels in eight out of nine instances. Additionally, AMOSA_MGMI_RW outperforms the others for problems with loose due-date tightness, and PAES_MGMI_RW works best for problems with moderate due-date tightness. Finally, for problems with strict due-date tightness, the MOSA_MGMI_FW surpasses the others for narrow due-date range, while PAES_MGMI_FW and PAES_MGMI_RW are better for moderate- to wide- due-date range.