Driving behavior is becoming more complex and personalized. It gradually become an active factor affecting traffic safety. In order to evaluate the safety risk level of driving behavior on urban roads, reduce traffic accidents. Against the traditional evaluation method is to give the person or vehicle factor unidirectionally. From the two aspects of driver operation safety and vehicle driving condition safety. Seven evalution were estabilshed that including speed and engine coolant temperature. K-means algorithm is used to perform cluster analysis on the two dimensions, and the driving behavior types are divided. The weight values of various evalution factors obtained by using the combination model of analytic hierarchy process (AHP) and entropy weight method (EWM) and evaluate the target layer combined with the fuzzy comprehensive evaluation method, divide driving behavior into safe, risky, and dangerous. In order to obtain the driving state data, OBD is used to collect driving state data. Analyzing the incentives for behaviors with different security risk levels. Providing reasonable basis for relevant vehicle management units to select and evaluate drivers.