Abstract: Large-scale mineral development has caused regional environmental and resource damage, geotechnical body location and morphology change, thus triggering geological disasters such as ground subsidence and collapse, which seriously affects the economic and social development and ecological environment restoration in mining areas, and seriously threatens the life and property safety of the people.With the continuous development of artificial intelligence technology, artificial neural networks are widely used in the field of mining subsidence prediction.We propose a mining subsidence prediction method based on LSTM(Long Short-Term Memory)and GRU(Gated Recurrent Unit)neural networks, and take the monitoring data of the working face of a mine as an example.The monitoring data of a mine face is used as an example to establish a time series prediction model of LSTM,GRU and its combination(SUM)to realize the prediction of surface subsidence and horizontal deformation caused by mining in the mining area.The prediction results of LSTM,GRU and SUM show that: on the whole, for the prediction of surface subsidence, LSTM and SUM are better than GRU,and LSTM is optimal, and its Root Mean Square Error(RMSE)and RMSE are better than GRU,and LSTM is better than GRU,and LSTM is the best.Mean Square Error(RMSE)and Mean Absolute Percentage Error(MAPE)are 14 mm and 1.5%,respectively; for the prediction of the surface level movement value, LSTM and SUM are better than GRU,and LSTM is optimal, and its RMSE and MAPE are 25 mm and 6.9%,respectively; for single prediction, the stability of accuracy from high to low is SUM,LSTM,and GRU,and the maximum and minimum values of RMSE and MAPE are GRU.