Reinforcement-learning-based improved snake optimizer for real-world engineering problems

The Snake Optimizer (SO) is a meta-heuristic inspired by the mating behaviour of snakes. It has demonstrated its ability to solve real-world engineering problems. However, its fixed parameters and motion strategy lead to premature convergence and slow solution refinement. This article proposes an improved snake optimizer based on reinforcement learning called the QSO that uses Q-learning to adjust the SO parameters dynamically. The goal is to achieve a delicate balance between exploration and exploitation. To enhance the population diversity further, adaptive mutation strategies tailored to different stages are proposed, and individuals can explore the search space in many ways. In addition, the integration of Sequential Quadratic Programming (SQP) enhances convergence. The QSO is compared to 19 advanced algorithms from CEC2017 and CEC2022. The performance of the QSO is verified on eight engineering design problems. The experimental results show that the QSO performs better in terms of both convergence speed and solution quality.