A proximal policy optimization algorithm based on adaptive hierarchical particle swarm

This paper proposes an adaptive hierarchical particle swarm optimization enhanced proximal policy optimization (AHPSO-PPO) framework. This framework addresses key challenges in reinforcement learning (RL), particularly slow convergence and premature convergence to sub-optimal policies. Firstly, adaptive parameter tuning, hierarchical particle swarm optimization, and variational evolution strategies are introduced in AHPSO. Secondly, the network parameters in RL are used as the optimization objective of the intelligent algorithm to obtain the optimal parameters faster and better. Finally, the experimental verification in a simulation environment shows that the AHPSO-PPO algorithm proposed in this paper performs better than standard PPO and DDPG. It has a faster convergence speed and a certain degree of robustness to environmental randomness. This hybrid approach links gradient-based optimization with group intelligence, providing a scalable solution for real-world decision-making systems.