Abstract: The correlation dimension is an important indicator for quantitatively analyzing the fractal properties of complex networks. In the existing correlation dimension method, the high-dimensional node vectors will be reconstructed by ergodic random walks when the network embedded in Euclidean space. The walking process only considers neighbors and causes the node vectors to contain more similar node information, which is not conducive to recovery dynamic characteristics of complex systems. To solve this problem, a new correlation dimension method for complex networks is proposed. The concept of delay time of embedded space in chaos theory is introduced into complex networks to increase the distance of traveling objects and reduce the correlation between nodes in the node vector. And we define the complex network correlation sum in non-Euclidean space, which makes correlation dimension method can be directly used in the traditional metric space. We validate the method from three networks with fractal dimensions. The results show that the method is effective in estimating the dimensions of the dynamic system, and the dimensions tend to the true correlation dimension values.