The invention discloses a safety early-warning system for underground pipeline facilities of the power grid. The system comprises a contact neural network paved in the region of an underground pipeline facility of the power grid, an optical fiber sensor connected with the contact neural network, a pre-posed intelligent collection terminal connected with the optical fiber sensor, and an early-warning platform and a field sound-light alarm device connected with the early-warning platform, wherein the early-warning platform is connected with the pre-posed intelligent collection terminal and is used for converting an intrusion electric signal into an intrusion model, comparing the intrusion model with a preset intrusion model in a model library and judging whether the intrusion model is destructive intrusion or interference vibration and transmitting alarm information if the intrusion model is destructive intrusion. The system can be used for improving the accuracy of early-warning.