Prediction for ground settlement of deep excavations based on wavelet analysis

Deep excavation will cause ground settlement inevitably. The measured data of ground settlement are usually disturbed by construction and surrounding enviroment, and the validation is greatly affected because of the noise in the settlement data. Based on the large amount of data collected from deep excavations, a new model combining the wavelet analysis with the radial basis function (RBF) neural network is proposed to predict ground settlement. The wavelet analysis is used to denoise effectively the measured data, and the settlement curve close to the practical situation can be obtained and taken as the characteristic vector of the RBF neural input layer. A prediction model for the wavelet network (W-RBF) is formed to predict ground settlement based on rolling prediction. The results of case study show that the prediction performance of W-RBF model is significantly better than that by using raw data with noises. It has high prediction accuracy and is fit for modern information construction.