Surface subsidence prediction is needed to evaluate mechanical safety and make it possible to provide further operation of buildings located in subsided area. The process of surface subsidence is studied by means of adaptive model of short-term time sequence prediction. The aim of adaptive forecast method is to build self-adjusting mathematical models that can show time-varying conditions and give accurate evaluation of given time sequence elements. Such models are primarily used for short-term forecasting. An appropriate forecast model is chosen according to the comparison of forecast values with the full-scale data of control benchmark subsidence. The results of modelling analysis show that Holt’s model forecasting has the nearest equivalent values to real subsidence value. As a result of numerical modelling the worst position of a five-storey panel building in displacement trough and stress-strain behavior of a standard panel building are defined. Numerical modelling results are proved by survey data.