Abstract: With the development of data collection, transfer and the requirement of safety in the construction, the deflection monitoring of the deep excavation becomes an integral part of the design as well as construction. The neural network approach through Back Propagation (BP) model and Auto Regressive with eXtra Input Signal (ARX) model has been applied in the deep excavation deflection monitoring and prediction. The results indicate that both models can satisfactorily predict the construction deflections. However, from the mean square error point view, BP has a better prediction results compared with the predictions made by ARX model.

Key words: Deep Excavation Deflection Prediction, Back Propagation (BP) Model, Auto Regressive with eXtra Input Signal (ARX) Model, Learning Efficiency, Result Efficiency, Result Comparison