<p>To reduce the influence of intersymbol interference among multiple terminals in the communication network of rail transit vehicles on the security early warning effect for network-classified information, this paper proposes a dynamic early warning method based on Parallel Interference Cancellation (PIC) for rail transit vehicle communication network-classified information security. The proposed method adopts PIC technology to reduce multiple access interference in the communication network of rail transit vehicles by selectively reconstructing interference signals, accurately detecting classified information bits sent by each communication terminal, and introducing the reliability grouping strategy of first-level detection to reduce unnecessary interference reconstruction. We construct a deep learning-based dynamic early warning model for classified information security in the communication network of rail transit vehicles. The Gated Recurrent Unit (GRU) module is used to extract the time-series characteristics of the classified information. The Multilayer Perceptron (MLP) module performs nonlinear mapping on the classified information to detect abnormalities. It then converts the abnormal detection result into probabilities for different early warning categories using the softmax function in the output module, selecting the category with the highest probability as the final dynamic early warning result. The experimental results show that the proposed method can accurately identify and warn the potential leakage or intrusion risk of classified information, and accurately output the security warning level of classified information. Compared with the comparison methods, the proposed method achieves 97% accuracy of classified information detection, only 1.2% false alarm rate and only 95 ms warning time, which reduces the computational complexity by 45%, and can provide a strong guarantee for the security protection of the communication network of rail transit vehicles.</p>