Biased Multiobjective Optimization and Decomposition Algorithm.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Multiobjective optimization problems (MOPs) present challenges for evolutionary algorithms due to features like bias.
• Balancing exploration and exploitation is crucial for algorithms to handle these difficulties.
• Decomposition-based MOEAs break down MOPs into single-objective subproblems.

Purpose of the Study:

• To propose a novel hybrid algorithm that integrates Differential Evolution (DE) and Covariance Matrix Adaptation Evolution Strategy (CMA-ES) within a decomposition-based MOEA framework.
• To address the challenge of bias in MOPs.
• To improve the balance between exploration and exploitation in solving MOPs.

Main Methods:

• A decomposition-based MOEA framework was employed.
• Single-objective subproblems were clustered into groups.
• Covariance Matrix Adaptation Evolution Strategy (CMA-ES) was used to optimize one subproblem per group, while Differential Evolution (DE) optimized others.
• CMA-ES was reinitialized upon meeting stopping criteria to address new subproblems within the same group.

Main Results:

• A new set of multiobjective test problems with bias features was constructed.
• Extensive experimental studies demonstrated the algorithm's suitability for handling biased problems.
• The proposed hybrid approach showed effectiveness in dealing with MOPs characterized by bias.

Conclusions:

• The proposed hybrid DE-CMA-ES algorithm effectively addresses biased multiobjective optimization problems.
• The strategy of clustering subproblems and selectively applying CMA-ES and DE offers a computationally efficient approach.
• This research contributes a valuable method for improving MOEA performance on challenging MOP instances.