基于值分布深度强化学习的桥式吊车轨迹规划与跟踪控制OA

Aiming at the problem that the traditional trajectory planning method in the bridge crane is complex in mathematical analysis and difficult to adapt to the dynamic environment in the complex model,a distributional soft actor-critic with three refinements(DSACT)deep reinforcement learning algorithm is proposed based on the value function distribution for real-time trajectory planning of a bridge crane.Richer state-action value distribution information can be captured by the DSACT algorithm through the distributed value function,which provides more comprehensive representation learning for decision making.Three optimization strategies(expected value substituting;twin value distribution learning;and variance-based critic gradient adjusting)are combined into the DSACT algorithm,which significantly improves the convergence speed,stability and robustness of the model.A sliding mode tracking controller is designed to achieve accurate tracking of the planned trajectory and improve the anti-interference ability and dynamic response performance of the system.Simulation results show that compared with the classical SAC and DDPG algorithms,the DSACT algorithm performs better in terms of planning accuracy,convergence speed and control stability.Compared with the traditional PID controller,the sliding mode tracking controller has significant advantages in tracking accuracy,anti-interference ability and dynamic response.