In order to improve the prediction accuracy of short-term traffic flow effectively, a short-term traffic flow local prediction method based on a combined kernel function relevance vector machine (CKF-RVM) model was proposed.Firstly, the C-C method was used to realize phase space reconstruction.Secondly, the number of neighboring points was determined by use of Hannan-Quinn criteria.Then, the CKF-RVM model was constructed based on particle swarm optimization algorithm.Finally, validation and comparative analysis was carried out using inductive loop data measured from the north-south viaduct in Shanghai.The experimental results demonstrate that the prediction error and the equal coefficient of the proposed method are both superior to the contrastive method.The MAPEs of the proposed method are 29.2%, 47.5% and 59.5% lower than GKF-RVM model, GKF-SVM model and weighted first-order local prediction model, which can further improve the prediction accuracy of short-term traffic flow.