Solving Fuzzy Bi-Objective Unrelated Parallel Machine Scheduling Problems with Flexible Constraints on Due Dates

• Main Authors: Wei-Ta Chu, 朱韋達
• Other Authors: 徐旭昇
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/38555702721718212820

碩士 === 元智大學 === 工業工程與管理學系 === 99 === This thesis studies unrelated parallel machine scheduling problems (UPMSP) that consider uncertainty on job processing data and job due dates, and have two simultaneous maximization objectives – average satisfaction of job tardiness and average satisfaction of the number of tardy jobs. Both objectives are concerned with the manufacturer’s performance on customer delivery. When dealing with uncertainty, fuzzy set theory may provide an acceptable compromise between expressive and computational difficulties for modeling preference and uncertainty. In the study, two measures based on fuzzy set theory are used to assess the satisfaction level of both objectives: possibility measure (height) and area ratio. Two archived meta-heuristics are applied to solve this bi-objective UPMSP: simulated annealing (SA) and tabu search (TS). In SA, random-weight direction (RWD) and fix-weight direction (FWD) are incorporated into the SA framework. In TS, other than RWD and FWD, a Pareto-based method adopting nadir distance (ND) to guide the TS iteration. The three TS search methods use shortest processing time rule (SPT) to construct an initial solution for each TS iteration step. In addition, greedy randomized adaptive search procedure (GRASP) is also applied to create an initial solution for TS. An experiment was conducted using five test sets with two problem sizes, 100 (jobs) x 5 (machines) and 200 x 10, which were generated based on Lee and Pinedo (1997) and Saidi Mehrabad et al. (2009). Each test set is characterized by problem size and due date factors, and has five instances. The numerical results indicate the following: (1) TS with SPT-initial solution outperforms TS with GRASP-initial solution; (2) The satisfaction values calculated based on possibility measure are generally higher than those based on area ratio method, but their conclusions are consistent with each other; (3) TS-FWD and TS-RWD perform better in terms of hyper-volume and generational distance performance measures for problems with loose due dates; on the other hand, there is little difference among SA and TS with distinct weighting direction approaches; finally, TS-ND is inferior to the other algorithms in most instances.