A nonlinear robust control method based on distributed deep reinforcement learning

As industrial automation continues to demand higher levels of intelligence and safety, Human-Computer Interaction (HCI) has become increasingly important in bridge crane control. Through interactive interfaces, operators can monitor system status in real time and provide feedback, ensuring both operational flexibility and safety. This paper proposes a nonlinear robust control method based on distributed deep reinforcement learning, designed for bridge crane control systems subject to sudden external disturbances. First, by obtaining the return distribution of the value function, the agent can learn and adapt more efficiently to the bridge crane control tasks. Then, prioritized experience replay is used to update the online weight network with relevant historical data stored in the database, enhancing learning efficiency. In addition, adversarial training is employed to disrupt the original state sequences, simulating real-world disturbances to accomplish the bridge crane hoisting tasks under complex working conditions. Finally, the simulation results demonstrate that this method excels in position accuracy and swing suppression, while maintaining reliable control performance even under disturbances.