A noise-immune extreme learning machine for short-term traffic flow forecasting

Abstract

Traffic flow prediction is an essential foundation of intelligent traffic management, and its accuracy and timeliness are essential indicators for effective traffic diversion and alleviation of traffic congestion. Aiming at the nonlinear relationship affecting traffic flow forecasting effect, a noise-immune extreme learning machine is proposed for shortterm traffic flow forecasting, which takes advantage of the gravitational search algorithm to search for an optimal global solution and used an extreme learning machine to forecast traffic flow. Extreme Learning Machine algorithm has high learning efficiency and strong generalization ability, which is widely used in regression, classification, and feature learning problems. However, due to the random setting of the input weights and the parameters of the bias matrix, the accuracy is not high, and the generalization ability is not strong. Therefore, the gravitational search algorithm is used to optimize the input weights and bias matrix to improve the accuracy of the prediction model. Based on the experimental data of Amsterdam Ring Road, the mean square error and mean absolute percentage error of the optimized model is reduced, which proves the effectiveness of the optimization. The noise-immune extreme learning machine model demonstrated superior performance and high prediction accuracy and can be well used in short-term traffic flow prediction.