Abstract: Considering the uncertainty and randomness of PM_2.5 concentration sequence, an interval prediction model based on complementary ensemble empirical mode decomposition and optimized Elman neural network was proposed. First, the original PM_2.5 concentration sequence was decomposed by using the complementary ensemble empirical mode. They were reorganized into several sub-sequences with the obvious differences in complexity by the sample entropy method. Second, a prediction model based on the multi input single output Elman neural network was established for each sub-sequence, respectively. Based on the results of each sub-sequence prediction, an interval prediction model based on the multi input double output Elman neural network was established for the prediction of PM_2.5 concentration. Finally, a novel interval prediction evaluation index was introduced as the objective function to further improve the prediction performance. The weight β and the threshold b of the Elman neural network were optimized by using the mind evolution algorithm. Based on the monitoring data, which was taken from the campus of Beijing University of Technology, the proposed prediction model was verified for reliable and good interval prediction results. It can provide a method for PM_2.5 concentration prediction.