Abstract: In order to improve the prediction accuracy of mining subsidence in mining area，a integrated prediction model with the combination of the Kalman filter model and Elman neural network is proposed based on the autoregressive integrated moving average model （ARIMA）.Firstly，in view of the non-stationarity and complexity characteristics of the subsidence mining monitoring sequence，ARIMA model is able to stabilize the original sequence，so as to construct the prediction model of surface subsidence and serve as the equation of state of Kalman filter；then，the results of Elman neural network subsidence prediction is introduced as the measured value into the Kalman filter measurement equation to establish the integrated prediction model；finally，for the selection of noise variance Q and R，the error characteristics of ARIMA model and Elman network model are calculated，so as to calculate the value of noise Q and R.Comparison of the prediction accuracy of the integrated prediction model proposed in this paper and BP neural network model，Elman neural network model，Kalman filter model.The results show that the root-mean-square-error （RMSE）of the prediction values and actual measured data of the four models are 2.06，5.857 8，2.926 9，3.688 9 mm respectively，the relative error of the four models are 1.170 4％、3.050 2％、1.432 6％ and 1.908 4％，mean absolute errors of them are 1.886 7，10.703 9，2.329 4，2.807 6 mm.The study results indicated that the integrated prediction model can effectively reduce the accumulation of errors of the same nature caused by a single prediction mechanism，the prediction accuracy of the integrated model is superior to the ones of other three models，it is help for improving the prediction accuracy of mining subsidence in mining area.

Key words: Mining subsidence, Kalman filter, Autoregressive integrated moving average model, Elman neural network, Integrated prediction model, BP neural network