Multiple Populations for Multiple Objectives: A Coevolutionary Technique for Solving Multiobjective Optimization...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Traditional multiobjective evolutionary algorithms (MOEAs) face challenges in fitness assignment due to conflicting objectives.
• Existing methods struggle to efficiently handle the complexity of multiobjective optimization problems (MOPs).

Purpose of the Study:

• To propose a novel coevolutionary technique, multiple populations for multiple objectives (MPMO), for developing MOEAs.
• To address the fitness assignment difficulties in MOEAs by decoupling objectives.
• To introduce coevolutionary multiswarm particle swarm optimization (CMPSO) as an implementation of MPMO.

Main Methods:

• Developed the MPMO technique, assigning each population to a single objective.
• Implemented CMPSO by integrating particle swarm optimization (PSO) within the MPMO framework.
• Enhanced CMPSO with an external shared archive for information exchange and novel velocity update and archive update strategies.

Main Results:

• CMPSO demonstrated superior performance across various benchmark MOPs compared to state-of-the-art algorithms.
• The MPMO approach effectively simplified fitness assignment by isolating objectives.
• Novel designs in CMPSO facilitated faster Pareto front approximation and introduced diversity to avoid local optima.

Conclusions:

• The proposed MPMO technique offers a simple and effective strategy for solving MOPs.
• CMPSO, based on MPMO, shows significant advantages in performance and efficiency for multiobjective optimization.
• The findings suggest MPMO as a promising general technique for future MOEA development.