Abstract:This paper investigates the active defense guidance problem for the hypersonic vehicle. The active defense guidance problem of the hypersonic vehicle is always subject to the limitations of incomplete observation information and observation noise in target-interceptor-defender scenarios. To tackle this issue, this paper introduces a reinforcement learning algorithm and proposes a cooperative active defense guidance based on a convolutional deep Q-network algorithm. In view of the spatiotemporal continuity properties of hypersonic vehicles, a stacking mechanism is proposed to process the incomplete information. The mechanism utilizes temporal dimension extension to compensate for the lack of spatial motion state information. Based on this, the convolutional neural networks are further employed to perform feature extraction on the stacked information. Trained by the shaped continuous reward function, the deep Q-network relies on the extracted feature tensor to obtain guidance. Finally, numerical experiments are performed to demonstrate the performance and robustness of the proposed active defense guidance, comparing it with the documented method.