论文部分内容阅读
针对多智能体完全合作环境下学习速度慢及收敛效果不佳问题,提出了基于分布式强化学习的二阶段适应学习方法,依次实现了智能体对环境的适应以及系统内部的协作.在第一阶段,智能体间的强化学习相互独立,以快速适应状态空间环境为主;该阶段中引入对环境的适应性因子,当智能体学习的误差小于该值时,智能体达到了对坏境的较高适应度.第二阶段中智能体采用不同的学习率进行交替适应学习,经过智能体间学习率的调整,实现了智能体学习系统中慢者与快者间的适应,最终形成协作直至收敛.与经典算法仿真结果的比较表明了二阶段适应性学习算法的可行性与高效性.
Aiming at the problem of slow learning speed and poor convergence in multi-agent cooperative environment, this paper proposes a two-stage adaptive learning method based on distributed reinforcement learning, which in turn enables the adaptation of the agent to the environment and the collaboration within the system. In the first At the stage, the reinforcement learning between agents is independent, so as to adapt quickly to the state space environment. At this stage, the adaptability factor to the environment is introduced. When the error of agent learning is less than this value, Higher fitness.In the second stage, the agents adopt different learning rates to adapt to learning alternately. After adjusting the learning rate among agents, the adaptation between the slow learners and the learners in the learning system of the agent is realized, The comparison with the classical algorithm shows that the two-stage adaptive learning algorithm is feasible and efficient.