Generalized regression neural network-based prediction methods for overlying strata failure zone height

The protection of phreatic water resources and prevention of roof water inrush hazard require accurate prediction of the overlying strata failure zone height (OSFZH) in coal mine. Considering the insufficiency of the traditional empirical formula for predicting OSFZH caused by underground coal mining, relationships between OSFZH and influence indicators were mathematically fitted firstly, based on the collected field measured data from nationally representative coal mines, revealing that the mining thickness was the most important index influencing on OSFZH with a linear relationship. Subsequently, the multiple nonlinear regression (MNR) model and generalized regression neural network (GRNN) model were established respectively, and that had a better prediction accuracy than the traditional empirical formula. More critical, the multiple nonlinear regression and generalized regression neural network were coalesced to establish a merging prediction model, and the MNR-GRNN merging model had high prediction accuracy and strong generalization ability. Finally, the insufficiency of the traditional method was discussed, and a numerical simulation case in Shilawusu coal mine, revealing the whole process development of overlying strata failure zone, was applied to verify the prediction accuracy of neural network-based prediction methods.