Abstract: In order to overcome the disadvantages of low precision and poor reliability of the single nonlinear mining subsidence prediction model, by comparing the characteristics of mining subsidence and the advantages and disadvantages of nonlinear prediction model, four models with strong adaptability and complementary performance, which are AR model, GM model, Cubic exponential smoothing model and Kalman filtering model are selected. Based on the model error square and minimum fusion criterion, a multi-source heterogeneous deformation prediction model is constructed for mining subsidence dynamic prediction. The observed values were used to calculate the weight coefficients of the multi-source heterogeneous fusion model sinking and horizontal prediction model. And the predictive performance of the model is tested by using the multi-source heterogeneous fusion model; the results show that the multi-source heterogeneous fusion model has high precision and good reliability compared with these four single models.