DEVELOPMENT OF AN INTELLIGENT DEFORMATION MONITORING AND EARLY-WARNING FRAMEWORK FOR HYDRAULIC STRUCTURES USING GNSS, INSAR, AND MACHINE LEARNING

Abstract

Progressive settlement, horizontal displacement, differential movement, foundation deformation, and seasonal structural response are important indicators of the operational condition of dams and other hydraulic structures. Conventional monitoring instruments provide accurate observations at selected points but may not adequately represent spatial deformation across large structures, while satellite-based interferometric synthetic aperture radar offers broad-area coverage but measures displacement primarily along the radar line of sight and can be affected by atmospheric disturbance, temporal decorrelation, vegetation, and data gaps. This study develops an integrated intelligent framework for deformation monitoring and early warning by combining Global Navigation Satellite System observations, satellite InSAR time series, environmental and operational variables, and machine-learning prediction. GNSS data provide continuous three-dimensional displacement at critical control points, whereas InSAR measurements map spatial settlement and differential deformation across the structure and surrounding foundation. The datasets are temporally synchronized, geometrically transformed into common displacement components, filtered for outliers and atmospheric noise, and fused using uncertainty-based weighting. A hybrid gated recurrent unit-attention model is proposed to forecast short-term deformation by learning nonlinear relationships among previous displacement, reservoir level, water temperature, air temperature, rainfall, and time-dependent structural response. The framework also includes anomaly detection based on displacement magnitude, deformation velocity, acceleration, residual error, spatial inconsistency, and persistence of abnormal behaviour. A multilevel early- warning index classifies structural response into normal, attention, warning, and critical states. The methodology is demonstrated through a representative hydraulic-structure monitoring scenario rather than a certified dataset from a specific dam. The expected contribution is a transparent decision-support system that combines point-scale accuracy, spatial coverage, short-term prediction, and engineering interpretation. The proposed framework can support targeted inspection, additional instrumentation, operational adjustment, maintenance planning, and timely intervention before progressive deformation develops into a major safety problem.