Dynamic Modeling of Passenger Car Units for Roundabouts: A Novel Genetic Algorithm Optimization-Based Approach

Abstract

This paper introduces a novel genetic algorithm (GA)-based optimization model for estimating passenger car units (PCUs) at roundabouts using occupancy time, addressing limitations in the existing methods that overlook traffic heterogeneity. The GA-based model achieved dynamic PCU values with a mean absolute error (MAE) of 3%–6% and a maximum error of 9%–10%, ensuring model simplicity and adaptability through the bias–variance trade-off technique. By estimating PCU values for different vehicle classes and analyzing roundabout geometry and traffic parameters, the model demonstrated high prediction accuracy with R-squared values between 0.995 and 0.998. Overfitting and underfitting were addressed by splitting data into training and testing sets and monitoring MAE changes. The model precision, reliability, and robustness were validated using multiple performance metrics. Monte Carlo sensitivity analysis and single-input single-output analysis assessed the relative importance of input parameters.