Comparison of particle swarm optimization and genetic algorithms for shop floor scheduling in manufacturing

This research presents a detailed comparison of Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) for shop floor scheduling. The evaluation highlights their respective strengths and weaknesses across multiple performance metrics. The main objective is to identify which algorithm performs better in terms of makespan, total flow time, total tardiness, computational time, convergence rate, and scalability. Experiments were conducted using various job sizes and real-world datasets, including the Kaggle-based Flexible Job Shop Scheduling Problem (FJSP). This experimental setup allowed for a critical examination of the outcomes and validation of the initial hypothesis. One key finding is that GA consistently outperforms PSO in makespan optimization, especially as problem size increases. GA’s use of crossover and mutation enables exploration of a broader solution space, leading to improved job sequences and shorter completion times. This aligns with existing literature emphasizing GA’s strong global search ability in combinatorial optimization. PSO, however, performs competitively in smaller scheduling problems. Its rapid convergence helps it reach near-optimal solutions quickly. In terms of total flow time, GA again demonstrates an advantage by reducing cumulative completion times through effective sequence optimization. PSO’s faster convergence sometimes results in poorer local solutions as problem size grows. For instance, in the 50-job instance, GA achieves around a 5% improvement in flow time over PSO. A similar trade-off is observed in total tardiness, where GA minimizes tardy jobs more effectively, while PSO struggles in certain configurations. Nevertheless, PSO stands out in computational efficiency, making it more suitable for real-time scheduling. Its ability to provide quick decisions is valuable in dynamic environments with job arrivals, machine breakdowns, and disruptions. The Kaggle dataset results reinforce this trade-off: PSO outperforms GA in computational time while maintaining competitive tardiness reduction.