Objective Oil and gas pipelines affected by landslides often suffer from abnormal monitoring data, and existing early-warning models are overly simplified. As a result, they cannot effectively reflect both current comprehensive warning and predictive warning for landslides and pipelines.

Methods A comprehensive early-warning method integrating data processing and physical models is proposed. The 3-sigma criterion and a sliding window are used for anomaly detection, and the Kalman filtering algorithm is applied to fill missing values and standardize monitoring data. A Long Short-Term Memory (LSTM) network is then used to predict standardized surface displacement and additional compressive stress in the pipeline. Based on standardized data, landslide deformation, external triggering factors, and receptor indicators are integrated to improve the four-level two-dimensional/three-dimensional early-warning matrix model for landslide hazards, which is applied to a pipeline landslide case in Mianyang.

Results The results show that the combination of the 3-sigma criterion, sliding window, and Kalman filtering effectively removes outliers and fills missing data. The LSTM model shows excellent predictive performance for surface displacement and pipeline stress. The current comprehensive landslide warning level is yellow. The prediction indicates that after half a month, surface displacement and pipeline stress will reach 160.97 mm and −55.99 MPa, respectively; both remain within the yellow-warning range and do not reach the orange-warning threshold.

Conclusion The proposed method effectively reduces the risk of false alarms caused by abnormal data and provides a reference for multi-source information fusion in landslide early warning, helping to gain time for on-site emergency response decision-making.