基于深度强化学习算法的车辆行为决策研究

Research on vehicle behavior decision based on deep reinforcement learning algorithm

针对传统的DDPG算法训练时间长和收敛速度慢的问题,提出一种将引导式学习和优选经验回放机制相结合融入DDPG的算法。改进的DDPG算法在训练初期的动作输出为引导式学习和策略网络共同作用的结果,后期引导式学习不再参与控制。同时引入经验池分离技术,将优势经验样本和劣势经验样本分开存储和固定比例随机抽取。在TORCS平台上进行车辆决策测试,结果表明,改进后的DDPG离,提高算法效率。

Aiming at the problems of long training time and slow convergence speed of the traditional DDPG algorithm,a combination of guided learning and optimal experience replay mechanism is proposed to be incorporated into the DDPG algo-rithm.The action output of the improved DDPG algorithm in the early stage of training is the combined effect of guided learning and strategy network,and later guided learning no longer participates in control.At the same time,the experience pool separation technology is introduced to separate the advantage experience samples and disadvantage experience samples and randomly select them at a fixed ratio.Through vehicle decision testing on the TORCS platform,which shows that the improved DDPG algorithm can effectively reduce training time,improve effective driving distance and improve algorithm ef-ficiency.