Aggregate level calibration of sumo for heterogeneous traffic conditions at Sri Lankan signalised intersections

Abstract

SUMO (Simulation of Urban MObility) provides considerable capability for traffic studies in developing countries. However, by default it is calibrated for lane-disciplined, homogeneous traffic of Europe. This study developed an aggregate level calibration strategy to make SUMO suitable for heterogeneous traffic in signalized intersections. In this regard, two traffic signalized intersections in Sri Lanka, namely Piliyandala (four-legged) and Katubedda (three-legged), were studied through detailed geometric and signal timing information. Traffic data collected through videos on site comprised vehicular classification, discharge rates, and queue lengths in each approach for each signal cycle, which were then aggregated at 15-minute intervals. Nine vehicle types corresponding to local traffic mixture were introduced, along with iterative calibration of behavioral parameters to match observed traffic operation. Model validity in Piliyandala was evaluated using the mean absolute percentage error (MAPE) and Geoffrey E. Havers (GEH) statistic, with results indicating satisfactory agreement for classified traffic discharge counts and queue lengths, as reflected by MAPE values below 15% and GEH values below 5, with minor exceptions in complex shared-lane queues. Once satisfied, the same SUMO parameter set was used in Katubedda to forecast the flow and the queue length. Results show that in Katubedda, statistical tests satisfy the accepted thresholds without further calibration. This establishes the model transferability of SUMO in various intersections in areas with similar traffic patterns without resource intensive individual calibration.