Integrating Actor-Critic Reinforcement Learning With Evolutionary Algorithm for Multimodal Multiobjective...

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Multimodal multiobjective optimization problems (MMOPs) demand a balance between diversity and convergence.
• Traditional algorithms exhibit limited adaptability in environmental selection, hindering performance on diverse MMOPs.

Purpose of the Study:

• To enhance environmental selection adaptability in evolutionary algorithms for MMOPs.
• To introduce a novel approach integrating actor-critic reinforcement learning (RL) with evolutionary algorithms.

Main Methods:

• Developed an RL process to dynamically optimize niche size, balancing diversity and convergence preferences.
• Defined state (convergence/diversity measures), action (niche size adjustment), and reward (state improvement).
• Employed actor and critic neural networks for real-time online learning and adaptive niching.

Main Results:

• The proposed algorithm demonstrated superior performance against ten state-of-the-art methods across 48 benchmark problems and a real-world application.
• Achieved significant improvements in maintaining the balance between diversity and convergence.
• Showcased enhanced overall optimization effectiveness compared to existing algorithms.

Conclusions:

• The integration of actor-critic RL with evolutionary algorithms significantly improves environmental selection adaptability for MMOPs.
• Adaptive niching technology, combined with local convergence assessment, provides a comprehensive evaluation of optimization.
• The proposed method offers a robust and effective solution for complex optimization challenges.