基于改进分层DQN算法的智能体路径规划
PATH PLANNING FOR AGENT BASED ON IMPROVED LAYERED DQN ALGORITHM
杨尚志 1张刚 1陈跃华 1何小龙1
作者信息
- 1. 宁波大学海运学院 浙江宁波 315211
- 折叠
摘要
针对智能体使用DQN(Deep Q Network)算法进行路径规划时存在收敛速度慢、Q值难以准确描述动作好坏的问题,提出一种优化DQN模型结构的分层DQN算法.该算法建立的激励层和动作层叠加生成更为准确的Q值,用于选择最优动作,使整个网络的抗干扰能力更强.仿真结果表明,智能体使用分层DQN算法的收敛速度更快,从而验证了算法的有效性.
Abstract
In order to solve the problems that the convergence speed is slow and it is difficult for Q value to describe the action accurately when an agent uses DQN algorithm in the process of path planning,a layered DQN algorithm optimized by the model structure of DQN is proposed.The excitation layer and the action layer built by the algorithm were superimposed to generate a more accurate Q value,which was used to select the optimal action and make the anti-interference ability of the whole network stronger.The simulation results show that the agent using layered DQN algorithm has a faster convergence speed,thus verifying the feasibility and effectiveness of the algorithm.
关键词
分层DQN/神经网络/强化学习/路径规划Key words
Layered DQN/Neural network/Reinforcement learning/Path planning引用本文复制引用
基金项目
国家自然科学基金(51675286)
浙江省重点研发计划(2018C02G2070536)
出版年
2024
国家科技期刊平台
NSTL
万方数据