Improving Performance and Robustness of Particle Swarm Optimization Metaheuristic Algorithms for Ridesharing Systems Based on a Cooperative Coevolution Approach

Optimization of ridesharing systems poses challenges for the development of solvers due to a nonconvex discrete solution space and complex constraints. Over the past decade, many metaheuristic algorithms have been proposed to solve optimization problems in ridesharing systems. Performance, robustness and efficiency are three important issues in the development of metaheuristic algorithms for ridesharing systems. Cooperative coevolution is a potential approach to improving the performance, robustness, and efficiency of metaheuristic algorithms. However, studies on the application of cooperative coevolution to optimization problems in ridesharing systems remain limited, as most existing work focuses on problems with a continuous solution space. Metaheuristic algorithms can be combined with the cooperative coevolution approach to solve optimization problems. In this paper, we combine particle swarm optimization (PSO) and bare-bones particle swarm optimization (BBPSO) with cooperative coevolution to develop two metaheuristic algorithms for ridesharing systems: discrete cooperative coevolution-based particle swarm optimization (DCCPSO) and discrete cooperative coevolution-based bare-bones particle swarm optimization (DCCBBPSO). We conducted a comparative study to assess their effectiveness in terms of performance, robustness and efficiency based on the experimental results. The results indicate that the cooperative coevolution-based metaheuristic algorithms developed in this study outperform discrete PSO (DPSO), discrete BBPSO (DBBPSO), and many other existing metaheuristic algorithms for ridesharing systems in terms of performance and robustness. The results show that the DCCPSO algorithm and the DCCBBPSO algorithm outperform the other 16 metaheuristic algorithms in convergence speed (measured by the average number of generations to find the best solution) in most cases. However, the DCCPSO and the DCCBBPSO algorithms do not outperform all the other 16 metaheuristic algorithms in terms of runtime. This is due to the inherent complex structure of the CC approach.