A traffic parameter short-time prediction method based on empirical mode decomposition (EMD) and classification combination prediction in an abnormal state relates to the technical field of traffic information. The prediction method includes being combined with the data sequence process method of the EMD, solving unstable data sequence of traffic parameters in an abnormal state into a stable intrinsic mode function (IMF) with multi-scale features; constructing a filter bank based on EMD filtering characteristics, reorganizing the IMF into high-frequency filtering, medium-frequency filtering, and low-frequency filtering; according to different characteristics of the IMF of each group, performing predictions by using the grey theory, kalman filtering and auto regressive moving average (ARMA) model respectively; accumulating results of all the groups to generate real-time predicting results of the traffic parameters of next time interval; and according to the real-time predicting results of the traffic parameters and historical data in the abnormal state, and performing multistep prediction so as to obtain a final predicting result of the traffic parameters and a future development tendency. The traffic parameter short-time prediction method based on the EMD and the classification combination prediction in the abnormal state has a better predicting capacity on the traffic parameters in the abnormal state and a future variation tendency.