Risk identification of the high arch dam in strong seismic regions using multisource monitoring data fusion

Effective risk identification for high arch dams in earthquake-prone regions requires the integration of multiple sources of monitoring data. This study proposes a three-state (normal, slightly abnormal, and significantly abnormal) decision-level fusion framework based on microseismic (MS) activity and radial plumb-line displacement. MS evidence is constructed by combining the significance of b -value changes evaluated using the Utsu test with return-period thresholds for event energy. Displacement evidence is derived from seasonal-trend decomposition using LOESS and isolation forest-based anomaly detection of the residual component. A deep neural network is used to map single-source features to state probabilities and generate basic probability assignments, which are subsequently fused using Dempster–Shafer evidence theory. Application to the Dagangshan high arch dam during the September 5, 2022, Luding earthquake ( M s = 6.8) reveals clustered high-energy MS activity and peak displacement residuals after the earthquake. The results indicate that the dam base exhibits the highest and most persistent risk, while the fusion framework reduces pre-event false alarms and provides zone-specific risk probabilities to support targeted inspection and maintenance.