Comparison of different calibration methods in modelling the unsignalized intersection using VISSIM with vehicular flow as the fitness measure

With the increasing attention towards microsimulation of transportation facilities, the calibration methods for simulation have been widened over the last decade, especially with advancements in computational power. Different methods of calibration exist in the literature. Hence, there is a need to compare different calibration methods for heterogeneous traffic conditions. Two unsignalized intersections (I-1 and I-2) were considered for the study. PTV VISSIM was used as the simulator and the intersections were modelled with data extracted from the field. The Vehicular flow was considered as the performance measure. Six driving behaviour parameters out of twelve were found to be sensitive in the model using the one-way ANOVA test. The range for each parameter was set. Three methods, namely manual calibration (trail & error method), automated calibration (Genetic Algorithm (GA) method) and partially manual calibration (trial & error and GA method), were used in calibrating the model. In manual calibration, 150 sets of parameters were sampled. The lowest Root Mean Square Normalised Error (RMSNE) value was found on both intersections after 1500 runs of simulation. The genetic algorithm was used with MATLAB for automated calibration and the models were calibrated through the Component Object Model (COM) interface. The objective function was defined and converged after the 49th and 32nd generation for intersections I-1 and I-2, respectively. The partially manual calibration used multiple regression models to find the optimal values for the parameters using the GA toolbox in MATLAB. The calibrated results from each method were compared with their merits and demerits.