A Comparison of Freeway Work Zone Capacity Prediction Models

To keep the freeway networks in a good condition, road works such as maintenance and reconstruction are carried out regularly. The resulting work zones including the related traffic management measures, give different traffic capacities of the infrastructures, which determines the travel time for road users. A work zone capacity prediction model therefore is highly needed to evaluate mobility. Considering the work zone capacity as a function of work zone configurations, different prediction models have been developed in the past. The conventional models assume a linear relationship between the capacity of a work zone and its configuration variables. Recent artificial intelligence models are more flexible in constructing nonlinear relationships, but the accuracy of the models is not suffiently tested. This research gives a comparison study of the existing models. Firstly, a selection of the critical work zone configuration variables is shortly discussed. Then three currently used prediction models are introduced, namely the model in the Highway Capacity Manual (2000), two multi-linear regression models, and a fuzzy logic based artificial neural network model. These models are tested for Dutch cases. Results show that comparing to the widely-applied linear regression models, the neuro-fuzzy model has the highest average accuracy and the prediction error can be reduced as large as 20%. The neuro-fuzzy model is recommended to serve in practice, as the choice of work zone configuration and the corresponding traffic measures can be made based on the capacity calculatio