Optimizing Multi-Objective Scheduling Problems of Unrelated Parallel Machines via Archived Metaheuristics with Matching Techniques

• Main Authors: Wei-Shung Chang, 張煒嵩
• Other Authors: 徐旭昇
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/69877375596789678864

博士 === 元智大學 === 工業工程與管理學系 === 99 === Over the years, parallel machine scheduling problems with a single objective or weighted sum of selected objectives have been widely studied. In practice, the layout of unrelated parallel machines is more common than that of their identical counterparts in real manufacturing environments, and management concerns of production scheduling are often multi-fold. In addition, there have been relatively few studies on unrelated parallel machine scheduling problems (UPMSP) considering multiple objectives and uncertain production situation. The research area of multi-objective unrelated machine scheduling problems (MO-UPSMP) is fertile, not only in development of theories and problem solving techniques, but also in practical applications. The main purpose of this thesis is to develop new algorithms for MO-UPMSP, and demonstrate that they are more effective than several popular algorithms. The thesis presents four studies on MO-UPMSP with sequence- and machine-dependent setup times. The first two studies present different evolutionary algorithms with matching-based decoding scheme to solve UPMSP with two minimization objectives: total weighted tardiness and total weighted flow time. In the two studies, the performances of the proposed hybrid evolutionary algorithms are compared with other metaheuristics, such as multi-objective simulated annealing (MOSA), and two well-known multi-objective population-based evolutionary algorithms. The third study aims to solve UPMSP with two configurations of three objectives: (1) three min-max objectives; (2) one min-max and two weighted-sum objectives. Three algorithms with matching-based decoding schemes are employed to solve the tri-objective UPMSPs, including SPEA2, CEMA (developed in this research), and GRASP. The GRASP has been successfully applied to solve many single-objective combinatorial optimization problems, but relative few research articles are published on multi-objective scheduling problems. We develop two matching-based decoding schemes, one for three min-max objectives and the other for mixed-type objectives. The experimental results indicate that CEMA with matching-based decoding is superior to the others in terms of several performance metrics often used in multi-objective optimization theory. The fourth study applies fuzzy approach to solve MO-UPMSP involving uncertainties on job processing times and job due dates. Two objectives are to maximize the decision maker’s satisfaction grade of production makespan and average satisfaction grade of jobs’ tardiness. The solution methods include archived MOSA and GRASP with path-relinking. The numerical results indicate that the GRASP technique has potentials to find good quality solutions for the problem under study.