Use of Two Penalty Values in Multiobjective Evolutionary Algorithm Based on Decomposition.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• The multiobjective evolutionary algorithm based on decomposition with the penalty-based boundary intersection (MOEA/D-PBI) is widely used.
• Appropriately specifying the penalty parameter in MOEA/D-PBI is challenging due to its impact on search behavior.
• Different penalty parameter values affect convergence and solution diversity differently.

Purpose of the Study:

• To address the challenge of penalty parameter specification in MOEA/D-PBI.
• To propose a novel method that leverages both convergence and diversification properties of the penalty-based boundary intersection function.
• To improve the performance of MOEA/D-PBI without sacrificing convergence speed.

Main Methods:

• Proposing the simultaneous use of two distinct penalty parameter values within the MOEA/D-PBI framework.
• Developing an algorithm that integrates small penalty parameters for convergence and large penalty parameters for diversity.
• Evaluating the proposed algorithm across a diverse set of test problems.

Main Results:

• The proposed algorithm effectively utilizes both small and large penalty parameter values.
• It achieves a balance between solution convergence and diversity.
• Experimental results show strong performance on various test problems.

Conclusions:

• Simultaneously employing two penalty parameter values in MOEA/D-PBI is a simple yet effective strategy.
• This approach enhances the algorithm's ability to achieve both convergence and diversification.
• The proposed method offers a promising improvement for MOEA/D-PBI applications.