With the development of science and technology and social progress, the landing of automatic driving technology seems to be close at hand, but the reality is that high-level automatic driving vehicles want to drive safely on urban roads, but also need to have the ability to accurately identify abnormal behaviors and accurately assess the risks in their operation. This paper focuses on the scientific issues in the process of abnormal behavior identification and risk assessment of autonomous driving formation vehicles in urban road environments. We constructed a framework for identifying abnormal behaviors of autonomous driving multi-vehicle formations in urban road environments, simulated urban road traffic scenarios using the CARLA-SUMO joint simulation platform, and completed several groups of independent simulation tests, and created a 17-dimensional multi-vehicle formation dataset by using five typical machine learning methods, and completed the identification and classification of six types of abnormal behaviors: sharp acceleration, sharp deceleration, overspeed, sharp steering, discrete, and falling out of line, abnormal behavior identification and classification.