Resumen
As a unique practice in travel demand modeling, traffic and revenue forecasting for toll facilities require highly accurate traffic projection. Explicit and rigorous statistical recognition of uncertainty in toll studies will mitigate the risk in planning and designing the toll facilities and better assist for future investments. Current traffic and revenue forecasting practice usually employs sensitivity analysis with alternative assumptions and Monte Carlo simulation to recognize these uncertainties. However, risk analysis using Monte Carlo simulation needs a large number of model runs and cannot directly use the sensitivity analysis results. This study proposes a neural network approach for toll road risk analysis. This approach incorporates the sensitivity analysis results into an effective forecasting procedure and develops the future revenue ranges with confidence intervals. Empirical results from a case study of I-75/SR 826 Express Lane project in Miami, Florida are presented. A comparison of Monte Carlo simulation results using the neural network model and the linear regression model suggests that the former would yield a more accurate, narrower range.