Abstract: Settlement deformation has always been a key problem in filling projects. The presence of wet-set loess and pulverized clay in the original foundation leads to increased and uneven settlement.Therefore, settlement monitoring and post-work prediction are essential to ensure the stability of high filling foundations. Based on the settlement monitoring data of high fill in an airport relocation project in northwest China, this paper applies hyperbolic, logarithmic and exponential fitting curves to predict the settlement of the fill foundation, analyzing and comparing them, and summarizes the fitting characteristics of each model. A settlement prediction model is established for high filling foundations using the gray system theory GM (1, 1), while a joint GM (1, 1)-BP neural network prediction model is proposed aimed at solving the problem of the deviation of the GM (1, 1) gray model from the measured curve in later stage. Finally the authors combine the gray GM (1, 1) model and the curve model with unequal weight coefficients in order to maximize the prediction accuracy of the GM (1, 1) gray model.The study provides reference for the prediction of similar foundation settlement in northwest China.