Mobile communication applications provide a unique data source for the research of human mobility pattern. Based on the distribution data of urban mobile phone users, in this paper is explored the macroscopic dynamical behavior of urban mobility human by using the method of random matrix theory. The largest eigenvalue and the corresponding eigenvector of mobile phone user data deviate far from the distribution of random matrix. The deviations from random matrix vary with time. We find that the largest eigenvalue corresponds to a whole behavior common to all urban human mobility. The results indicate the temporal trends of the mean of correlation coefficient and the largest eigenvalue. We also find that the spatio temporal evolution of the weight of eigenvector components for the eigenvector corresponding to the largest eigenvalue is very consistent with the fluctuation pattern of the macroscopic behavior of urban human mobility.