A Reinforcement Learning-Based Bi-Population Nutcracker Optimizer for Global Optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• The traditional nutcracker optimizer algorithm (NOA) faces challenges in balancing global exploration and local exploitation, often leading to local optima.
• Complex optimization problems require algorithms with robust mechanisms for both exploration and exploitation.

Purpose of the Study:

• To introduce a novel reinforcement learning-based bi-population nutcracker optimizer algorithm (RLNOA).
• To enhance the performance of the NOA in solving complex optimization problems by improving the balance between global exploration and local exploitation.

Main Methods:

• A bi-population mechanism divides the population into exploration and exploitation sub-populations based on fitness.
• An improved foraging strategy using random opposition-based learning enhances diversity in the exploration sub-population.
• Q-learning is employed as an adaptive selector for exploitation strategies in the exploitation sub-population.

Main Results:

• The RLNOA demonstrated superior performance compared to nine state-of-the-art metaheuristic algorithms.
• Evaluations on CEC-2014, CEC-2017, and CEC-2020 benchmark function sets validated the algorithm's effectiveness.
• The proposed RLNOA effectively balances global exploration and local exploitation, mitigating local optima entrapment.

Conclusions:

• The RLNOA significantly improves upon the traditional NOA by effectively addressing the exploration-exploitation dilemma.
• The integration of reinforcement learning and a bi-population strategy offers a powerful approach for complex optimization tasks.
• The RLNOA presents a promising advancement in metaheuristic optimization algorithms.