Abstract: In view of the complexity and nonlinear characteristics of the prediction results，a new prediction model of surface deformation in mining areas is constructed based on the generalized regression neural network(GRNN).Firstly，the modeling principles of GRNN are discussed and the key factors that affect the prediction accuracy of GRNN model are introduced.Then，in order to improve the generalization ability and prediction accuracy of the network，the network is modeled and trained by adopting the rolling modeling method.The optimal smoothing factor SPREAD is determined in line with the across validation algorithm based on RMSE.Finally，the optimized GRNN is applied to predict the surface deformation in a mining area.The prediction results of BP neural l network based on Levenberg-Maquardt algorithm，RBF neural network and regression analysis method are used to compare with one of the optimized GRNN.The results show that，the GRNN net work generalization ability and prediction accuracy are better than the others，in addition，the algorithm of optimized GRNN is stable.So，the optimized GRNN is suitable for surface deformation prediction in mining areas.

Key words: Surface deformation in mining areas, Generalized regression neural network(GRNN) , Rolling modeling, Across validation, Prediction