REAL-TIME GNSS DISPLACEMENT COARSE DIFFERENCE DETECTION SEQUENCES FOR LANDSLIDE MONITORING

The Global Navigation Satellite System (GNSS) offers high-precision three-dimensional surface deformation data and has found extensive application in the monitoring and early warning of geological hazards such as landslides. However, in complex mountainous terrains, GNSS displacement monitoring data can be marred by gross errors due to factors like hardware malfunctions, data communication issues, and challenging observation conditions, which may not accurately reflect the true deformation patterns. To tackle this issue, this paper proposes and implements a real-time anomaly detection method for GNSS displacement data based on Robust Random Cut Forest (RRCF). The algorithm was validated and analyzed using both simulated and actual measurement data. The results indicate that under ideal prior model conditions, the RRCF method slightly outperforms the 3σ wavelet-method with slightly lower accuracy and precision rates but a slightly higher recall rate, yielding an overall anomaly detection evaluation index of 0.9757. Conversely, under non-ideal prior model simulations, the RRCF method is slightly less effective than the 3σ wavelet-method, with slightly lower accuracy, precision, and recall rates, resulting in an overall evaluation index of 0.9768. For the real-time anomaly detection scenario, the method was further verified using actual displacement monitoring data from geological hazards. The results revealed that the RRCF method effectively identifies abnormal abrupt changes in GNSS displacement data, aligning well with the occurrence of actual outliers and maintaining a low false positive rate. It demonstrates high detection accuracy and reliability, providing technical support for the Landslide real-time monitoring and early warning of GNSS displacements.