Short-term Traffic Flow Prediction Based on Improved Deep Echo State Network

Abstract

In order to improve the processing capability of deep echo networks for short-time traffic flow prediction problems, an improved deep echo state network (IDESN) is proposed in this paper. The improved deep echo state network algorithm firstly improves the activation function in the traditional echo state network and uses a particle swarm algorithm to optimize the parameters in the new activation function. Secondly, the circular greedy algorithm is used to find the hyperparameters of the improved deep echo state network. Finally, the wavelet threshold denoising algorithm is used to denoise the traffic flow sequences. In this paper, three short-term traffic flow datasets are used for testing. The results show that the MSE values of the three datasets are reduced by 57.13%, 57.80% and 51.59%, respectively, compared with the original deep echo state network as well as the improved deep echo state network has higher accuracy.