Queue Length Prediction at Un-Signalized Intersections with Heterogeneous Traffic Conditions

Abstract

Increasing queue lengths while reducing average vehicle speeds is a notable criterion in intersections with heterogeneous traffic conditions. Such queue lengths vary with different intersection controls. Thisstudy aimed to estimate the queue length at un-signalized intersections with heterogeneous traffic conditions. The study was done for un-signalized intersections in Peradeniya and Weliwita, Sri Lanka and the data were collected through video recordings. The queue lengths in an un-signalized intersection with mixed traffic conditions have an instantaneous aggressive variation due to the uncontrolled movements. Thus, a time series analysis with the aid of Vector Auto Regression (VAR) model was used in order to estimate the queue length. Variables considered in this study were arrival flow rate, discharge flow rate, number of conflicts for 15 seconds time intervals as independent variables and queue length at the end of each 15 seconds as the dependent variable. For the modelling, the procedure of “Box-Jenkins” method was followed. After the confirmation of the variables are stationary, Cointegration check and Granger causality tests were done to check the cointegration between variables and the granger causality between variables. Then, VAR models were developed using 80% data from the total data set for both locations. The remaining 20% of the data set was used to validate the model using the MAE, MAPE, and RMSE error values between the actual and predicted queues. Among both models, 0.94 of higher R2 value and Durbin Watson value as 2 was obtained for the developed model using raw variables for Weliwita junction. Furthermore, the observed MAE, MAPE, and RMSE values for Weliwita model were 3,5 and 6%, respectively. Thus, the results of this study can be used to reduce traffic congestion while enhancing the safety of the users at un-signalized intersections in Sri Lanka.