Sensitivity of Signal Control Performance to Prediction Errors in Traffic Flow Data

Abstract

The traffic flow data used for optimizing signal parameters is inherently inaccurate because it is usually determined by prediction. In order to examine the sensitivity of control performance to the prediction errors, the total delays for signal control strategies based on erroneous flow data are calculated when they are applied to the actual flow. A discrete-time model is developed to describe the dynamic behavior of oversaturated urban road networks. This model enables to optimize a series of splits for each signal that minimize the total network delay during congested periods. Several numerical examples for a two-way street consisted of five intersections are given and the results are discussed. The control strategy using predicted flow data in optimization produces 10% to 23% more delay than the true optimum control. The total delay is found to be sensitive to errors in traffic demand or proportion of traffic movements, while not to be sensitive to errors in saturation flow. The results obtained in this paper should be useful to estimate the range of conditions for which particular signal settings are likely to be suitable.