| Summary: | 碩士 === 國立清華大學 === 工業工程與工程管理學系 === 97 === Abstract
This study presents methods to solve single project, non-preemptive infinite mode RCPSP where resources are renewable and the intensity of each activity is fixed within the activity duration. The objective of the problem is to minimize project makespan. A piece-wise linear curve is proposed to approximate the continuous, nonlinear relations between the duration and the intensity of an activity. Such a project scheduling problem that uses piece-wise linear curves is called approximated problem in this study.
Enumerative approach (Wu, 2007) generates all the possible event sequences to obtain optimal solution for approximated problems. However, the solution time of enumerative approach is too long when the problem size is relatively large, so only small size problem can be solved within the limited computation time. Thus, we use heuristic methods to solve relatively large problems. The heuristics suggested in this study can be divided into two stages. The stage 1 includes three methods -- multiple-run Boctor approach (Wang, 2008), activity-based beam search approach and event-based beam search approach. In stage 1, the three heuristic methods will be used to efficiently find good event sequences. The stage 2 includes two heuristic search methods -- tabu search and simulated annealing. In stage 2, the two search algorithms further improve the solutions obtained in stage 1 by searching for a better event sequence.
In stage 1, the experimental results show that event-based beam search approach is the best heuristic method to efficiently find a good event sequence among the three heuristic methods. In stage 2, the results show that simulated annealing performs better than tabu search within the same CPU computation time.
Keywords: project scheduling; linear programming; beam search; tabu search; simulated annealing
|
|---|
