Abstract: Aiming at the difficulty in short-term accurate prediction of flight delay, a combined flight delay prediction model based on improved denoising method of ICEEMDAN and support vector machine (SVM) is built. Firstly, the original flight delay sequence is decomposed into stationary components by using ICEEMDAN algorithm, and then the mixed noise in the component is determined by correlation function analysis and is processed by SG filtering. Secondly, the SVM regression prediction model is established for each component according to its features; predicted values of each Component model are superimposed to obtain the final predicted data. Compared with the ICEEMDAN-SVM model, the improved combined prediction model reduces the root mean square error and mean absolute percentage error by 8.7% and 11.9% respectively, proving that the model has good following ability to flight delay sequence fluctuation and strong generalization capacity.

Key words: flight delay short-term prediction, EMD, SG filter wave, SVM