Abstract:
The use of time models for subsidence prediction is one of the most commonly methods for predicting dynamic deformation in coal mining areas. Based on analyzing the existence of time-zero problems in typical time function models, this paper introduces a Forest Growth Model (i.e., the Hossfeld model), and analyzes the prediction accuracy of the Hossfeld model for the subsidence prediction at a single point and an arbitrary points in the coal mining subsidence basin by using the leveling data and the D-InSAR results, and evaluates the correlation of the model parameters. The results show that in the prediction results of single-point subsidence from leveling data, the accuracy of Knothe and Usher models with corrected time-zero is higher than that of Knothe and Usher models with uncorrected time-zero; the proportion of the subsidence prediction results with the RMSE<100 mm from the Hossfeld model is slightly lower than that of the Usher model with corrected time-zero, higher than that of the Usher model with uncorrected time-zero, and much higher than that of the Knothe time function model with corrected and uncorrected time-zero; the Hossfeld model has the highest accuracy in the MAE <100 mm; in the prediction results of arbitrary points subsidence from D-InSAR technology observation, the statistical correlations of the parameters used to construct the dynamic prediction model show that the Hossfeldmodel is the most coherent; further, the calculation of the Bland-Altman shows that the difference between the results of the Hossfeld model and the D-InSAR results is relatively small and has the highest accuracy at RMSE and MAE <20 mm. Compared with Knothe model and Usher model, the Hossfeld model has obvious advantages in accuracy for dynamic subsidence prediction because it does not require time-zero correction.