Abstract: The dynamic prediction of mining surface subsidence describes the relationship between surface movement deformation and mining time, which has important guiding significance in the prevention and control of surface movement deformation and failure in goaf.Taking the measured data of 1414（1） fully mechanized mining face in Naner mining area under the conditions of thick loose layer mining in a coal mine in Huainan as an example, six time functions such as Knothe, segmented Knothe, Weibull, Gompertz, Logistic and Usher were used to construct the dynamic prediction model of surface subsidence in the working face.Based on the Matlab curve fitting toolbox, the model parameters were solved, and the prediction accuracy of each model was evaluated by mean square error and relative mean square error.The results show that the models with relatively good accuracy and stability are the piecewise Knothe, Gompertz and Usher function models.The Gompertz model is the best, and the error and relative mean square error are 42.6 mm and 4.6%, respectively.The Knothe model is significantly worse than the other five models; the Usher model can realize single point prediction and whole basin surface dynamic prediction, and can achieve the optimal prediction effect at individual points.