Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, İnşaat Mühendisliği Bölümü, Türkiye
Tezin Onay Tarihi: 2011
Tezin Dili: İngilizce
Öğrenci: Erdem Özleyen
Danışman: RİFAT SÖNMEZ
Özet:The resource-constrained project scheduling problem (RCPSP) aims to find a schedule of minimum makespan by starting each activity such that resource constraints and precedence constraints are respected. However, as the problem is NP-hard (Non-Deterministic Polynomial-Time Hard) in the strong sense, the performance of exact procedures is limited and can only solve small-sized project networks. In this study a genetic algorithm is proposed for the RCPSP. The proposed genetic algorithm (GA) aims to find near-optimal solutions and also overcomes the poor performance of the exact procedures for large-sized project networks. Contrarily to a traditional GA, the proposed algorithm employs two independent populations: left population that consist of leftjustified (forward) schedules and right population that consist of right-justified (backward) schedules. The repeated cycle updates the left (right) population by maintaining it with transformed right (left) individuals. By doing so, the algorithm uses two different scheduling characteristics. Moreover, the algorithm provides a new two-point crossover operator that selects the parents according to their resource requirement mechanism. The algorithm also includes a modified mutation operator which just accepts the improved solutions. Experiment results show that the suggested algorithm outperforms the well known commercial software packages; Primavera Project Planner (P6 version 7.0) and Microsoft Project 2010 for the RCPSP. In addition, the algorithm is tested with problems obtained from literature as well as the benchmark PSPLIB (Project Scheduling Problem Library) problems. The proposed algorithm obtained satisfactory results especially for the problems with 120 and 300 activities. Limitations of the proposed genetic algorithm are addressed and possible further studies are advised.