Real Time Road Performance Prediction Using Multilayer Perceptron Fusion of ETC and Connected Vehicle Data

Connected vehicles (CV) are key components of cooperative intelligent transportation systems. As CV penetration rates and roadside communication coverage gradually increase, government agencies face growing infrastructure investment needs. By fusing CV with existing electronic toll collection (ETC) data, road performance prediction accuracy can be enhanced, potentially enabling better resource allocation. This study employs multilayer perceptron neural networks to fuse CV and ETC data to predict real time average speeds over freeway segments. Statistical regression is adopted to analyze the effect of variables, including penetration rate, coverage rate, data transmission frequency, crash response time, and demand level, on prediction accuracy. Training, validation, and testing data for the data fusion model are synthesized using a calibrated simulation model. The results show that when CV communication coverage is below 100%, the fusion of CV and ETC data achieves higher prediction accuracy than using either data source alone. As CV penetration rates increase, improving communication coverage gradually becomes more critical. Finally, logistic regression is developed to provide guidance for management agencies to determine the appropriate timing for implementing the data fusion approach.