An Optimization-Based Iterative Learning Method for Anticipatory Network Traffic Control

Anticipatory traffic control determines signal timings to optimize network performance, taking into account travelers’ route choice responses and the resulting flow patterns. In general, the route choice response is predicted through traffic assignment models. However, model-reality mismatch usually brings suboptimal conditions characterized by unexpected congestion to the real-life system. The objective of this paper is first to propose an iterative optimizing control algorithm to elevate traffic system to its real optimal performance, by learning from errors between modeled and measured flows. A key algorithmic implementation issue regarding calculation of the network flow sensitivity is then addressed. We present a method to estimate the derivative of real flows with respect to control variables. The proposed algorithm is tested on a simple network as well as a midsize network. Numerical examples confirm the effectiveness of the new reality-tracking control strategy and its ability to identify (local) optimal solutions on real traffic networks.