Adaptive Real-Time Prediction Model for Short-Term Traffic Flow Uncertainty

Abstract

In order to promote the accuracy of short-term traffic flow forecasting, an adaptive real-time model consisting of two important stages is proposed. The first stage encloses a novel online sequence extreme learning machine with forgetting factor (FFOS-ELM) that effectively averts the influence of early data on model accuracy induced by the time variability of short-term traffic flow and adaptively corrects the model parameters. In the second stage, based on the optimal estimation on the particle filter system, optimized real-time forecasting of future traffic volume is accomplished by filtering out the noise in the original traffic volume. Finally, the validity and feasibility of the proposed model are verified by a case study. Microwave data from the main road of a city in China was selected to extract the traffic volume as the model data set, and the accuracy of the proposed model is compared with five traditional offline algorithm models and two online algorithm models. Forecasting results indicate that the two-stage adaptive model produces more accurate and stable predictions and shows potential in forecasting the short-term traffic flow under uncontainable conditions.