A correct understanding of the evolution characteristics of the extinction coefficient series under different humidity conditions is the premise and foundation to establish the correction model of atmospheric particulate matter humidity. Using the data of mass concentration of PM2.5 (excluding rainy days), together with the homologous surface visibility and relative humidity data from June 2013 to May 2014 at 4th Section, South Renmin Road of Chengdu, the corresponding unit mass extinction coefficient time series is retrieved. The complex evolution of the extinction coefficient during moisture absorption is briefly discussed, and the non-universality of the existing humidity correction model is exemplified. Based on the reconstruction of phase space theories, the optimal delay time f and the embedding dimension m are determined, and several characteristic quantities, including the saturation correlation dimension, the maximum Lyapunov index and the Komogorove entropy, are also calculated. The results show that this series has the characteristics of low dimensional chaos. The Cao method is applied to excluding the possibility that the unit mass extinction coefficient time series is nonlinear. By adopting the surrogate data method, the time series is finally proved to be stochastic. The results not only clarify the characteristics of the unit mass extinction coefficient series, but also establish the theoretical foundation for the improvement of the atmospheric particulate matter humidity.