[Purpose]To address the issue in China's current specifications whose rutting model uses a single function for prediction, leading to low estimation accuracy and poor applicability, a three-stage rutting model for asphalt pavement was proposed based on long-term rut measure data, and MATLAB was employed to develop a data-driven model algorithm. [methods]Through the accelerated loading test of full-scale pavement facility, the three-stage rutting model and its algorithm were constructed and verified by four asphalt pavement structures including semi-rigid base asphalt pavement, rigid base asphalt pavement, flexible base asphalt pavement and full-thickness asphalt pavement after 70 million equivalent loadings. The rutting models of China Specifications for Highway Design of Asphalt Pavement and the MEPDG of US were comparatively studied. [result] there are significant differences among different pavement structures, and the three-stage rutting model can accurately characterize the various stages of asphalt pavement rutting development of all four pavement structures. The developed model algorithm can realize intelligent judgment of the three-stage rutting development, and the fitting effect is good; MEPDG-2015 rutting model has a bad prediction for the middle and later stages of rutting, and the D50 rutting model has poor applicability to rigid bases asphalt pavement and full-thickness asphalt pavement structures. The existing rutting models use one single function, which is not sufficient to accurately characterize the multi-stage development of asphalt pavement rutting. [conclusion]The research results can improve the prediction of asphalt pavement rutting, and provide a reference for the application of big data analysis in asphalt pavement performance prediction and maintenance decision-making.