Optimal, integrated and adaptive traffic corridor control: A machine learning approach

The objective of this research approach is to develop a self-learning, adaptive, independent and integrated freeway-control for both recurring and non-recurring congestion. This research project illustrates an innovative Artificial Intelligence approach as an effective methodology for traffic corridor control. This approach provides a method for real-time control of traffic on freeways and corridors with express/collector or freeway/arterial network systems under both recurring and non-recurring congestion.; Advancement in Intelligent Transportation Systems (ITS) and communication technology has the potential to considerably reduce delay and congestion through an array of network-wide strategies. The two most promising control tools are traffic responsive ramp metering and/or traffic diversion, using variable message signs (VMS). Technically, the use of these control methods independently might limit their potential usefulness. Therefore, integrated corridor control using ramp metering and VMS diversion methods simultaneously can be synergic and beneficial. This dissertation focuses on corridor control using ramp metering and VMS routing independently and in an integrated manner.; The research introduces the use of Reinforcement Learning (RL), an Artificial Intelligence method for machine learning, so as to provide a single, multiple or integrated optimal control agent for a freeway or freeway-arterial or express/collector corridor for both recurring and non-recurring congestion. Reinforcement learning is an approach whereby the control agent directly learns optimal strategies via feedback reward signals from its environment. A simple but powerful reinforcement learning method known as Q-learning is used. A micro simulation tool - PARAMICS has been utilized to train the agent in an offline mode within a simulation environment in order to make it ready for field implementation.; Even though this approach has been proven effective and superior in many fields, there have been very few attempts to use it in traffic control. The approach developed in this thesis is rigorously evaluated, using various networks in different combinations of control strategies under simulated conditions. Results from these simulation case studies in the Toronto area are very encouraging and have demonstrated the effectiveness and superiority of the technique in reducing congestion.