Abstract: Achieving precise dynamic prediction of surface subsidence is a key technology for ensuring the safety of sur face structures in mining-affected areas and optimizing maintenance plans.The prediction accuracy of dynamic subsidence fore casting algorithms combining the probability integral method and dynamic time-series functions often fails to meet the require ments for maintenance and management.The primary reason for this phenomenon is the variation in parameters of dynamic time-series functions due to the complexity of mining subsidence.To address this issue,this paper proposes a high-precision dy namic prediction method based on the concept of "mechanism-constrained data-driven" modeling.The probability integral method is adopted as the physical framework for subsidence calculation,and the Knothe time function is used to describe the temporal evolution of surface point subsidence.Multi-period measured subsidence data are utilized to invert the optimal time parameter for each observation period using the Levenberg-Marquardt algorithm,thereby obtaining the dynamic evolution se quence of the time parameter.The metabolic grey model is then employed to predict future parameter values,enabling synchro nous dynamic updates of both the parameters and the predicted subsidence values,and consequently achieving high-precision dynamic surface subsidence prediction.Case studies were conducted,and the results show that compared with the method using fixed parameters in the dynamic time function,dynamically updating the time-series function parameters reduces the relative er ror of dynamic prediction from 25.1% to within 5%,significantly improving the prediction accuracy.In another case,the root mean square error in each period was reduced by more than 50% on average.This study provides a new technical approach for dynamicpredictionofminingsubsidenceandhas important implications forguidingengineeringpractice.

Key words: miningsubsidence,dynamicprediction,timeseries,probabilityintegralmethod,timefunctionparameters