Dynamic Evaluation of Aquifer Water Abundance Under Non-Stationary Conditions Based on TVP-CKF

Accurate prediction of aquifer water abundance is critical for coal mine safety, yet traditional static models often fail to capture the spatial heterogeneity and non-stationarity of hydrogeological conditions. This study proposes a dynamic evaluation methodology integrating Grey Relational Analysis, the Analytic Hierarchy Process, and a Time-Varying Parameter Cubature Kalman Filter (TVP-CKF). By reconceptualizing spatial borehole data as a dynamic time-series process, the model recursively updates the contribution weights of six controlling factors based on monitoring data from 2012 to 2020. Analysis reveals a structural shift in the groundwater system: the influence of hydrochemical factors (TDS) has diminished, while hydraulic conductivity has become the dominant control over time. The TVP-CKF model significantly outperformed static regression and recursive least squares baselines, demonstrating superior convergence stability and precisely capturing transient inflow fluctuations. Furthermore, its uncertainty quantification effectively bounded extreme low-flow events within 95% confidence intervals. This approach validates the necessity of adaptive modeling in evolving geological environments, providing a robust, risk-quantified tool for precise water inrush prevention.