Aiming at the long training time and slow convergence rate of BP neural network,wavelet basis consisted by the translation and stretch of mother wavelet constitutes the activation function of neural network basing on the theory of wavelet analysis.The neural network can use simple topology to approximate function by the guidance of network initialization and parameters selection.The nonlinear relationship between the bearing capacity of CFG piles composite foundation and its main factors is established by using the trained network.Under the same structure and parameters,the prediction results are analyzed and compared with the BP neural network.The result shows that this forecast model makes full use of the wavelet transformation to analysis the data of time-frequency localization and is combined with the self-learning function of artificial neural network,which makes it have a strong ability to approach and fault tolerance.The precision and speed by using the trained wavelet neural network is higher and its predicting result is more accurate than that of BP network.