Abstract: At present, people often predict soft-ground settlement with the method of exponential curve and double-curve extent; but the result is not ideal. The application of neural network to that exists some limitations. GM(1,1) model has been applied to this field, but all the equal-time settlement data model do not explain definitely interpolation method in existed case. The paper takes the predicting the soft ground settlement of Xibu tunnel filling sea project in Shenzhen for example, discusses the way to solve the problem through building the BP neural network-grey system united model; constructing the equal-time settlement time series data by BP neural network nonlinear interpolation method, on the basis of it, building the time series GM model for settlement, and building the time function to predict the settlement. The case study shows that the model is quite accurate in short-term settlement prediction, its long-term settlement prediction also has some project referential value.

Key words: soft ground, BP neural network, nonlinear interpolation, GM(1,1) model, settlement prediction