Abstract: In view of the poor adaptability of the traditional combination forecasting model to the dynamic prediction of mining subsidence, and the solution of combination weight does not take into account the influence of the freshness of the measured data and the low efficiency of the solution. Through the method of measurement analysis and theoretical modeling, the combination of data freshness prediction model is studied. The main conclusions are as follows: according to the principle of good stability, high prediction precision and good complementarity, the combined model selected GM （1, 1）, cubic exponential smoothing model and AR model as a single model. Introducing the freshness function Knothe, based on the minimum square error criterion, the optimal combination prediction model for mining subsidence is constructed for the first time considering the data freshness function Knothe. The genetic algorithm-nonlinear programming algorithm is used to obtain the weight of combined forecasting model. Experiments show that the freshness of optimal combinative forecasting model has obvious advantages in accuracy and reliability than the observation value of the freshness of the optimal combinative forecasting model which did not take into account, the combined model based on the variance reciprocal method, equal weight combined forecast model and the single prediction model.