The Short-term Prediction of Vehicle Speed of Urban Freeway Based on RBF Neural Network

Abstract

The short-term prediction of traffic flow is a significant way to help predicting and solving traffic congestion. In order to describe traffic state, speed is used as a parameter to find the prediction model. By analyzing the temporal and spatial relevance of speed, a speed prediction model is established based on temporal-spatial characteristics and radial basis function (RBF) neural network. With the simulation and analysis of microwave data of Second Ring in Beijing, the prediction result demonstrates that the prediction method based on temporal and spatial characteristics can effectively improve the prediction accuracy of speed compared with the prediction method based on a single time series, where the mean absolute relative error (MAPE) decreases from 7.45% to 3.61%.