Abstract: The aim of this research is to improve the precision of pit deformation prediction and enhance the stability of prediction results. Support vector machine, BP neural network and GM (1,1) are used as the basis of prediction model, and the corresponding first-order prediction models with parameters optimized are established. On this basis, the second-order combinatorial forecasting model of multiple fixed weight and non-fixed weight is established. In subsequence, on the basis of the Markov chain theory, the error correction model of three steps is established, and the progressive prediction of foundation pit deformation is realized. Results demonstrate that the prediction accuracy and stability are greatly improved by the progressive prediction of multiple stages, which verifies the validity and feasibility of the proposed method in this paper. The result is expected to provide a new idea for the prediction of foundation pit deformation.

Key words: foundation pit, progressive prediction model, support vector machines, BP neural network, GM(1, 1), combinatorial forecasting, error correction