We proposed a novel intelligent system for long-distance optical fiber pre-warning and investigated its predictive accuracy in real working environment. The system mainly includes two parts: distributed sensing and signal recognition. Specifically, the phase-sensitive optical time domain reflection (Φ-OTDR) technology was used for distributed sensing and the neural network technology was used for signal recognition so as to recognize and classify intrusion events. In addition, an improved neural network structure was proposed in the signal recognition and compared with a neural network based on wavelet packet decomposition and that with three hidden layers. Finally, we performed three different experiments to explore the recognition accuracy of the system. The results show that the proposed system can well recognize and classify the intrusion events with an average recognition rate above 95%.