Short time traffic flow identification and prediction at road junctions using Recurrent Multilayer Perceptron

摘要

Short time traffic forecasting is the prediction of traffic parameters like traffic flow into the future to avoid congestions and accidents. A variety of short-time traffic flow forecasting methods have been developed in the last two decades but most of the literatures focus on predicting traffic flow on a single road, not on junctions. However, Junctions are responsible for 80-90% of congestion and 40-60% of accidents on traffic systems. In this paper, Recurrent Multilayer Perceptron (RMLP) trained with Levenberg-Marquardt algorithm is used to forecast traffic flow at a junction found in Addis Ababa, Ethiopia; locally known as Teklehaimanot junction fifteen minute in to the future. The result is then compared with Multilayer Perceptron (MLP) trained with gradient descent back-propagation algorithm. Forecasting results show that RMLP trained with Levenberg-Marquardt algorithm on the junction has approximately three times better training Mean Square Error (MSE). In addition to this, the RMLP has a correlation value of 93.48% for a completely new set of validation data but the MLP's correlation value for the same new set of validation data is 82.9% when used to predict traffic flow on the junction.