The prediction of ground surface settlement due to excavation of a deep foundation pit is a com- plex and non-linear problem

in which there are many influential factors of a highly non-linear relation- ship.The traditional theory of settlement prediction has some limitations and need further improvement on its accuracy in practical application.The artificial neural network is a non-linear dynamic system of multi- ple variables

and can conveniently and flexibly simulate any unknown system of complex polygene so that the ground surface settlement around a deep foundation pit can be predicted with all main influential fac- tors being taken into account properly.This article introduces the module

learning process

improved al- gorithm of the back propagation（BP）network

as well as the module with its learning process of the radi- al basis function（RBF）network.We analyze the major factors influencing the ground surface settlement due to excavation of the deep foundation pit.Twenty-five samples of actual measurements of ground sur- face settlement around deep foundation pits have been taken for the training of the neural networks to build 11 models of BP neural network and RBP neural network with the imported factors.The training process and the predicting precision of five validating samples show the feasibility and accuracy of the artificial neu- ral network for the prediction of ground surface settlement around deep foundation pits.