Abstract:

In order to improve the prediction accuracy of foundation pit deformation and rationally evaluate the safety state of the foundation pit, a combined prediction model is established, in which the support vector machine, extreme learning machine and GM (1,1) model are proposed as single prediction model and the combination weight value is determined by fixed weight method and nonfixed weight method. The safety evaluation indices of the foundation pit deformation are constructed by using cumulative deformation and deformation control value to evaluate the safety state of the foundation pit. The standard deviation method is used to analyze the development trend of foundation pit safety. The case analytical results show that: (1) The combined prediction is superior to single prediction in terms of prediction accuracy, prediction risk and prediction results stability. (2) The prediction accuracy of nonfixed weight method is superior to that of fixed weight method, and the combination effect of BP neural network weight method is the optimum. (3) The prediction results show that the foundation pit is in dangerous and necessary safety measures should be conducted; and the prediction results coincide with the analytical results, which illustrates the effectiveness and feasibility of the prediction and evaluation methods mentioned above.

Key words: foundation pit, combined prediction, fixed weight method, nonfixed weight method, safety, trend analysis