Group-based approach to predictive delay model based on incremental queue accumulations for adaptive traffic control systems

In this study, we develop a mathematical framework to estimate lane-based incremental queue accumulations with group-based variables and a predictive model of lane-based control delay. Our objective is to establish the rolling horizon approach to lane-based control delay for group-based optimization of signal timings in adaptive traffic control systems. The challenges involved in this task include identification of the most appropriate incremental queue accumulations based on group-based variables for individual lanes to the queueing formation patterns and establishment of the rolling horizon procedure for predicting the future components of lane-based incremental queue accumulations in the time windows. For lane-based estimation of incremental queue accumulations, temporal and spatial information were collected on the basis of estimated lane-based queue lengths from our previous research to estimate lane-based incremental queue accumulations. We interpret the given signal plan as group-based variables, including the start and duration of the effective green time and the cycle time. Adjustment factors are defined to identify the characteristics of the control delay in a specific cycle and to clarify the relationship between group-based variables and the temporal information of queue lengths in the proposed estimation method. We construct the rolling horizon procedure based on Kalman filters with appropriate time windows. Lane-based queue lengths at an inflection point and adjustment factors in the previous cycle are used to estimate the adjustment factors, arrival rates, and discharge rates in the next cycle, in which the predictive computation is performed in the current cycle. In the simulations sets and the case study, the proposed model is robust and accurate for estimation of lane-based control delay under a wide range of traffic conditions. Adjustment factors play a significant role in increasing the accuracy of the proposed model and in classifying queueing patterns in a specific cycle. The Kalman filters enhance the accuracy of the predictions by minimizing the error terms caused by the fluctuation in traffic flow.