Channel state information (CSI)-based physical layer authentication (PLA) is considered to be a promising technology for wireless network security. However, in mobile scenarios, device mobility and environmental dynamics limit the reliability of PLA due to channel state instability and spatial location uncertainty. To improve authentication reliability, we propose a knowledge-enhanced PLA algorithm. This algorithm leverages prior knowledge of mobile devices and wireless channels to verify mobility patterns and extract stable features. Specifically, trajectory verification, exploiting the space-time constraints of the mobile device, is conducted before and after location prediction to assess the rationality of the mobile trajectory. Moreover, a three-scale feature model is proposed to extract stable features for precise location prediction, which models the stable characteristics of unstable mobile channels from different sensitivities to spatial variations. Evaluation results show that the proposed algorithm can achieve higher reliability than two state-of-the-art algorithms in three mobile scenarios.