Adapting Decomposed Directions for Evolutionary Multiobjective Optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Decomposition methods are common in evolutionary algorithms for multiobjective optimization problems (MOPs).
• Existing methods often struggle with complex Pareto fronts (PFs) due to reliance on single ideal/nadir points.
• This limitation hinders effective exploration of diverse PFs.

Purpose of the Study:

• To propose an effective method, adapted decomposed directions (ADDs), for solving MOPs.
• To address the limitations of single-point guidance in decomposition-based MOP algorithms.
• To enhance the coverage and accuracy of Pareto front approximation.

Main Methods:

• Developed a novel decomposition approach using exclusive ideal points for each weight vector.
• Adapted decomposed directions dynamically during the evolutionary search process.
• Integrated ADDs into three representative multiobjective evolutionary algorithms (MOEAs).

Main Results:

• Theoretically analyzed the effectiveness of the proposed ADDs method.
• Experimentally verified significant performance improvements when ADDs were embedded in MOEAs.
• ADDs demonstrated superior performance compared to seven competitive MOEAs across 39 artificial and 1 real-world MOPs.

Conclusions:

• The proposed ADDs method effectively addresses MOPs with challenging Pareto front shapes.
• Adapted decomposed directions ensure even and complete coverage of the Pareto front.
• ADDs offer a robust enhancement for multiobjective evolutionary algorithms.