Abstract: According to a series of features such as nonlinear,non-stationary,multi-scale exhibited from composite foundation settlement time sequences,a composite foundation settlement prediction model,combined of empirical mode decomposition(EMD)and least squares support vector machine(LS-SVM),is proposed.By EMD decomposition,the non-stationary sequence of the settlement originating from observational data is decomposed into several different dimensions smooth IMF(intrinsic mode components)components,of which each is predicted by the LS-SVM,which will eventually combine those components to reach the forecast results about the original settlement data sequence.Verified in the practical application of engineering example,the method can not only obtain high prediction accuracy,but reflect the mutation of the original settlement data,both of the two indicating that this method is of promotional value.