Traffic flow prediction model based on improved variational mode decomposition and error correction

Abstract

With the aggravation of traffic congestion, traffic flow data (TFD) prediction is very important for traffic managers to control traffic congestion and for traffic participants to plan their trips. However, its effective prediction faces great difficulties and challenges. Aiming at handling complexity of TFD, a new TFD prediction model based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), neural network estimation time entropy (NNetEn), variational mode decomposition (VMD) improved by northern goshawk optimization (NGO) algorithm, kernel extreme learning machine (KELM) improved by artificial rabbits optimization (ARO) algorithm and error correction (EC) is proposed. Aiming at choosing the decomposition layers and penalty coefficient of VMD, VMD improved by NGO, named NVMD, is proposed. Aiming at handling the problem of selecting KELM parameters, KELM improved by ARO, ARO-KELM, is proposed. Firstly, CEEMDAN is used to decompose TFD into a limited number of IMF components. NNetEn is used to divide IMF components into high- and low-complexity components. The sum of high-complexity components is selected for secondary decomposition by NVMD. Then ARO-KELM is used to predict all decomposed components. Finally, error correction is introduced to further improve the prediction accuracy. TFD from England highway is used in the experiments. Taking TFD I as an example, the RMSE, MAE, MAPE and R2 are 4.5682, 3.3104, 0.0458 and 0. 9997 respectively. The results show that the proposed model is superior to the other six comparison models at 99% confidence level, which provides a theoretical and data basis for controlling traffic jams, accidents and pollution.