Long-distance oil and gas pipelines are widely distributed and have complex background environments. Therefore, their optical-fiber pre-warning system experiences a high false-alarm rate in identifying destructive events that threaten pipeline safety in a real-world environment. This makes it challenging for the system to achieve accurate pre-warning results and ensure pipeline safety. This study applies deep learning to a long-distance fiber pre-warning system. Through deep learning, a vehicle-passing signal that mainly affects the pre-warning effect is identified, which effectively reduces the false-alarm rate of the pre-warning system. The intelligent fiber pre-warning system is mainly divided into two parts: the distributed optical-fiber sensing system and the signal-recognition system. In a real-world environment, an intrusion signal around the pipeline is collected by a Φ-OTDR(phase-sensitive optical time domain reflectometry) distributed optical-fiber sensing system. Additionally, a recognition model is established by convolutional long short-term memory and fully connected deep neural networks to detect the vehicle-passing signal. After training and blind testing, the vehicle-passing event recognition model demonstrated a good recognition and positioning effect in a real-world long-distance fiber-monitoring environment and effectively reduced the false positives of the pre-warning system.