关键词: 开采沉陷　, 非稳沉数据　, 时间函数　, 遗传算法　, 动态预计　, 概率积分模型

Abstract: Dynamic prediction of coal mining subsidence is significant for surface damage assessment and land reclamation.
Current mainstream methods construct dynamic models by integrating maximum subsidence values with time functions,but
their parameters predominantly rely on existing data from working faces with similar geological conditions or are inverted using
leveling monitoring data after subsidence stabilization. To address the time-lag problem in subsidence prediction caused by parameter
inversion dependent on stable subsidence data when lacking analogous geological parameters,this study proposes a dynamic
mining subsidence prediction method based on non-steady subsidence data and Genetic Algorithm (GA),termed DSPGAUD
(Dynamic Subsidence Prediction via Genetic Algorithm with Unstable Data). The method first inverts parameters using
non-steady subsidence sample data during mining operations through GA algorithm to build a single-point dynamic subsidence
model. It then selects and couples models based on single-point predictions,and finally establishes a global dynamic prediction
model by integrating the probability integral method. Validated with a working face in Zhuxianzhuang Mine,Northern Anhui
Province,results demonstrate that:① Compared to conventional methods,DSP-GAUD improves timeliness by over 51%,while
reducing the average fRMSE and fMAE of early-stage predictions by 68%;② The coupled dynamic prediction model synergizes ad
vantages of multiple time functions,achieving superior accuracy (mean R2 =0. 96) to single-function models;③ The proposed
model for the emergence time (t1) of maximum subsidence velocity versus cutting-face distance exhibits stronger adaptability
than stable-subsidence-data-based models. This method significantly enhances prediction timeliness and early-stage accuracy,
offering certain practical value for real-time mining subsidence prediction.

Key words: mining subsidence,non-steady subsidence data,time function,genetic algorithm,dynamic prediction,probability
integral mode