Abstract：To improve convergence speed and accuracy of short-term traffic flow prediction, a method of KPCA-KELM combining kernel principal component analysis (KPCA) and kernel extreme learning machine (KELM) was proposed. By the KPCA method, the nonlinear principal elements of model input in the feature space were effectively extracted to realize the data pre-processing. In the KELM method, the nodes number of network hidden layer was not set, and the output weight of network was calculated by the regularized least squares algorithm to achieve good promotion with extremely fast learning speed. The advantages of KPCA and KELM were combined in the proposed method. The traffic flow prediction data measured by the ITS Research Group of the University of Washington in Seattle and the Beijing Traffic Administration were tested, and the KPCA-KELM method was compared with single KELM, LSSVM, SVM, KPCA-LSSVM, KPCA-SVM and other prediction methods. The experimental results show that the convergence speed and prediction accuracy of the proposed method are better than those of other comparison methods. For the single-step prediction of the measured traffic volume data of Beijing Transportation Administration, the prediction accuracy of the KPCA-KELM method is 1.991 3 higher than that of the KELM method.
2018, Vol. 39 
