Flerobjektiv Resursberänsad Projektschemaläggning: En Jämförelsestudie av Metaheuristiska Metoder : Att Balansera Kostnad, Tid och Resurser i Schemaläggning av Telekomprojekt

Background. Real-world projects often need to be optimized in several objectives simultaneously while adhering to resource constraints. The multi-objective resource-constrained project scheduling problem (MO-RCPSP) models these trade-offs but is computationally difficult to solve exactly as project size increases (in theoretical computer science it is defined as a NP-hard-problem). This motivates using metaheuristic approaches that generate diverse schedules which explore the different optimal objective-trade-offs (Pareto-optimal solutions). Objective. A comparative evaluation of three metaheuristic algorithms: Multi-Objective Harris Hawks Optimization (MOHHO), Particle Swarm Optimization (MOPSO) and Ant Colony Optimization (MOACO), alongside the control algorithms; NSGAII and SPEA2. The algorithms will optimize schedules across small, medium and large scheduling instances with regard to cost, duration and resource utilization. Method. A unified tri-objective formulation was implemented together with three metaheuristic algorithms. Hyperparameter tuning was performed via grid search on a medium-sized instance under a 30-second time budget. The optimal configurations found were then used for the main experiment. In the main experiment each algorithm executed ten independent 30-second trials per instance, maintaining an archive of non-dominated solutions. Performance was assessed with the metaheuristic performance metrics: normalized hypervolume, generational distance, spread and coverage, supported by three-dimensional Pareto visualizations. Result. On small instances (10 tasks), MOPSO achieved the highest normalized hypervolume (0.737) and best diversity. On medium instances (50 tasks), NSGAII led in convergence and coverage (NHV = 0.719, coverage = 0.734). On a large real-world instance (147 tasks), NSGAII remained dominant (NHV = 0.670) while MOHHO achieved the best generational distance, indicating strong convergence. MOACO underperformed at all scales. Conclusion. Metaheuristic performance depends on problem scale. Swarm-based strategies excel on small problems, elitist evolutionary methods perform best at medium scale and adaptive exploration–exploitation mechanisms become advantageous in large, high-dimensional scheduling. Algorithm selection should align with project complexity and computational constraints.