A new simplified MOPSO based on Swarm Elitism and Swarm Memory: MO-ETPSO

Ricardo Fitas

2024-02-20

Abstract:This paper presents an algorithm based on Particle Swarm Optimization (PSO), adapted for multi-objective optimization problems: the Elitist PSO (MO-ETPSO). The proposed algorithm integrates core strategies from the well-established NSGA-II approach, such as the Crowding Distance Algorithm, while leveraging the advantages of Swarm Intelligence in terms of individual and social cognition. A novel aspect of the algorithm is the introduction of a swarm memory and swarm elitism, which may turn the adoption of NSGA-II strategies in PSO. These features enhance the algorithm's ability to retain and utilize high-quality solutions throughout optimization. Furthermore, all operators within the algorithm are intentionally designed for simplicity, ensuring ease of replication and implementation in various settings. Preliminary comparisons with the NSGA-II algorithm for the Green Vehicle Routing Problem, both in terms of solutions found and convergence, have yielded promising results in favor of MO-ETPSO.

Neural and Evolutionary Computing

What problem does this paper attempt to address?

This paper proposes a simplified multi-objective particle swarm optimization algorithm (MO-ETPSO) based on group elitism and group memory. The algorithm combines core strategies from the NSGA-II algorithm, such as the crowding distance algorithm, and leverages the advantages of individual and social cognition in swarm intelligence. The novelty of the algorithm lies in the introduction of group memory and group elitism, which enables the adoption of NSGA-II strategies in particle swarm optimization. These features enhance the algorithm's ability to preserve and utilize high-quality solutions throughout the optimization process.

Preliminary comparisons were made between MO-ETPSO and NSGA-II in the green vehicle routing problem, and the results showed that MO-ETPSO has advantages in terms of the solutions found and convergence. The paper introduces the fundamentals of optimization problems, multi-objective optimization, Pareto dominance, and scalarization methods through the definition and explanation. The authors also discuss different optimization techniques, including genetic algorithms, ant colony optimization, and particle swarm optimization, with specific mention of NSGA-II and MOPSO algorithms.

In MO-ETPSO, group memory is used to store information about the best solutions, while group elitism ensures that the optimal solutions are preserved over several generations. The algorithm evaluates the final value of particles through their ranking, crowding distance calculation, and constraint penalization. In this way, it considers not only solutions close to the Pareto frontier but also maintains the diversity of solutions.

The paper demonstrates the effectiveness of MO-ETPSO through a numerical example of the green vehicle routing problem (GVRP) and compares it with NSGA-II. The experimental setup and results show that MO-ETPSO can effectively operate in high-dimensional search spaces and outperforms NSGA-II under certain conditions.

In summary, the objective of the paper is to propose a simplified version of the multi-objective particle swarm optimization algorithm, which combines the strategies of NSGA-II while simplifying the implementation and demonstrating its superiority in practical problems.