PROJECT TITLE :

Prediction of Traffic Flow via Connnected Vehicles

ABSTRACT:

We propose a framework for short-term traffic flow prediction (STP) so that transportation authorities can take early actions to control flow and prevent congestion in the future. We make predictions regarding flow on a specific road segment based on historical flow data and cutting-edge features such as real-time feeds and trajectory data provided by connected vehicles (CV) technology. These predictions are based on the time frames that will occur in the future. We demonstrate how this novel approach enables advanced modeling by integrating into the flow forecasting, the impact of the various events that CV realistically encountered on segments along their trajectory. This is done so as a means of compensating for the fact that existing approaches do not adapt to variations in the volume of traffic. We solve the STP problem by training a Deep Neural Network (DNN) to perform multiple tasks simultaneously while incorporating input from CV. According to the findings, our method, specifically MTL-CV, which has an average Root-Mean-Square Error (RMSE) of 0.052, outperforms state-of-the-art ARIMA time series, which has an RMSE of 0.255, as well as baseline classifiers (RMSE of 0.122). When compared to MTL-CV, the performance of ANN was lower, with an RMSE of 0.113. This was the case when comparing ANN to single task learning with an artificial neural network (ANN). MTL-CV was able to learn historical similarities between segments as opposed to directly using historical trends in the measure. This is because trends may not exist in the measure, but they do exist in the similarities.