An Efficient Dynamic Resource Allocation Framework for Evolutionary Bilevel Optimization.

Area of Science:

• Computational mathematics
• Optimization theory
• Evolutionary computation

Background:

• Bilevel optimization problems (BLOPs) involve hierarchical decision-making.
• Evolutionary algorithms (EAs) are common for BLOPs but resource-intensive.
• Existing EAs struggle with redundant lower-level iterations, hindering efficiency.

Purpose of the Study:

• To propose an efficient dynamic resource allocation framework for evolutionary bilevel optimization.
• To address the significant resource consumption and redundant computations in solving BLOPs.
• To enhance the convergence and efficiency of EAs for complex bilevel problems.

Main Methods:

• Introduced DRC-BLEA, a dynamic resource allocation framework.
• Implemented a competitive quasi-parallel paradigm for lower-level tasks.
• Utilized continuously updated selection probabilities and a cooperation mechanism.

Main Results:

• DRC-BLEA achieves competitive accuracy across diverse problem sets.
• Significantly reduces the number of function evaluations compared to state-of-the-art methods.
• Demonstrates a substantial decrease in overall running time.

Conclusions:

• DRC-BLEA offers an effective solution for efficient evolutionary bilevel optimization.
• The proposed framework overcomes resource limitations and redundant computations.
• DRC-BLEA shows strong performance in both benchmark and real-world scenarios.