In this paper, an online model-free approach based on the largest Lyapunov exponents (LLE) and angular velocity deviation is provided for coherent groups identification (CGI). Firstly, by using a model-free LLE calculation method, the generator coherency identification (GCI) criterions are proposed by establishing the mathematical relationship between the LLE and angular velocity difference. Secondly, an online CGI scheme for power systems is designed. Compared with the existing measurement-based methods, the breakthrough work of this paper is that the proposed GCI criterions combines the stability theory of nonlinear dynamic trajectory based on LLE with the physical mechanism of the out-of-step between generators, such that it can improve both the speed and the accuracy of the GCI. Besides, the proposed approach needs not to find the optimal observation window size and shorten the required observation window size greatly. Moreover, it can be applied to online CGI with small computation requirement. Extensive test results on New England 39-Bus System and practical East China (EC) power grid, as well as the comparisons with existing CGI methods verify high accuracy and efficiency of the proposed approach. INDEX TERMS Coherency identification, largest Lyapunov exponents (LLE), dynamic features, post-fault phase trajectory, wide-area measurement system (WAMS).