Network Traffic Prediction of Echo State Network Based on Dropout

• 1、Beijing University of Posts and Telecommunications,School of Cyberspace Security

Abstract：In this paper，an echo state network model based on dropout method (dropout ESN) is proposed. The model avoids the randomness of the reservoir generated by the classical echo state network, reduces the connectivity of neurons in the reservoir, optimizes the structure of the reservoir, solves the over fitting problem in the process of network traffic prediction, and improves the prediction efficiency of the network. The echo state network based on dropout method is applied to the actual network traffic prediction task. The neurons in the reservoir of dropout ESN are set to stop working with different probability. The classic echo state network and the echo state network based on dropout method are compared, and the influence of the two algorithms on the prediction performance is analyzed.

Keywords： neural network echo state network network traffic prediction dropout