A Rough-to-Fine Evolutionary Multiobjective Optimization Algorithm.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Many real-world problems involve multiobjective optimization, where solutions are often initially distant from the desired Pareto-optimal set.
• Existing algorithms may struggle with efficiency when dealing with solutions far from the Pareto front.

Purpose of the Study:

• To propose a rough-to-fine evolutionary multiobjective optimization algorithm.
• To address challenges in optimizing problems with initial solutions far from the Pareto front.
• To improve the efficiency and effectiveness of multiobjective optimization.

Main Methods:

• A decomposition-based evolutionary algorithm is developed.
• A modified k-means algorithm constructs a subproblem tree using uniform weight vectors.
• Subproblems are solved level-by-level, refining the approximation of the Pareto front.

Main Results:

• The algorithm successfully approaches the Pareto front by progressively refining solutions.
• It demonstrates effectiveness for problems where initial solutions are far from the Pareto set.
• Theoretical analysis indicates a lower time complexity of O(M logN) for handling new candidate solutions.

Conclusions:

• The proposed rough-to-fine algorithm offers an efficient strategy for multiobjective optimization.
• Its level-by-level refinement approach is advantageous for problems with distant initial solutions.
• Empirical evidence supports the algorithm's efficacy and improved time complexity.