To address the problem that the increased positioning errors in wireless positioning for intelligent vehicles caused by non-line-of-sight (NLOS) signals, a robust UWB/IMU fusion positioning methodology based on reliable identification of NLOS signals is proposed. Firstly, the coarse NLOS identification is conducted based on a support vector machine （SVM） learning model and a multi-sensor consistency mathematical model respectively. Subsequently, the fine NLOS identification model based on D-S evidence theory is designed to effectively integrate the results of the aforementioned models at the decision level. Finally, a multi-sensor adaptive fusion positioning method based on factor graph is proposed to dynamically adjust the fusion model according to the results of NLOS identification, in order to achieve robust positioning for intelligent vehicles in NLOS environments. The results of real vehicle experiments indicate that, in terms of NLOS identification performance, compared with the conventional SVM model, the proposed method improves the precision, recall and accuracy by 6.97%, 5.37% and 6.36% respectively. In terms of positioning performance, compared with the existing conventional least squares positioning method, the proposed method reduces the root mean square error, the maximum error, and the standard deviation by 12.55%, 63.40%, and 13.23%, respectively, effectively improving the positioning accuracy and robustness of intelligent vehicles in NLOS environments, and overcoming the shortcomings of traditional methods in low positioning accuracy and poor reliability in NLOS environments.