Probabilistic analysis of signalized intersection delay and level of service

A deterministic procedure is provided in Highway Capacity Manual (HCM) 2000 to estimate control delay and LOS for signalized intersections with a single value for each input variable. However, the single values of major input variables have inherent variation and uncertainty due to the stochastic nature of traffic. Small variations in each input variable may result in a relatively large uncertainty in estimated delay.; Five uncertainty analysis methods were selected for potential application in traffic analysis: First Order Analysis (FOA), Monte Carlo Simulation (MCS), Fourier Amplitude Sensitivity Test (FAST), Point Estimate Method (PEM), and Fuzzy Set Theory (FST). Based on the preliminary comparison three methods, FOA, MCS, and PEM, were selected for further investigation with varied saturation levels and uncertainty levels of inputs. It was also shown that probability distributions have a slight effect on the uncertainty of delay and correlations can significantly reduce the uncertainty of delay. Three types of sensitivity analysis methods were investigated: local methods (OAAT and PDA), regression-based methods (SRC and PCC) and variance decomposition methods (method of Sobol and EFAST). The investigation of uncertainty and sensitivity analysis methods suggests that MCS and SRC always produce accurate results and work well together. They are recommended for comprehensive analysis of the uncertainty of delay. PEM and OAAT provide acceptable result and require less computation. They could be applied when only a simple examination is desired.; A procedure for probabilistic analysis was proposed which features a combination of uncertainty and sensitivity analysis and an iterative application based on uncertainty analysis results and desired accuracy. This research also initiates a prototype database for uncertainty of inputs with three sets of data: current field data, NCHRP data and other literature data. With the probabilistic method, outputs change from a single delay value and LOS grade, to mean value, standard deviation, confidence interval and probability density function for the delay, and likelihood percentages for each applicable LOS grade such as C(10%), D(76%), and E(14%). Theses provide users a comprehensive view of the LOS of a signalized intersection and are more realistic and robust than that from the current method.