engleski

Application of Q-learning to ramp metering in cases of significant changes in traffic demand

The number of traffic participants has significantly increased in recent decades. Such a trend is also evident in road traffic. The result is that today’s urban highways are under influence of the now increased traffic demand and cannot fulfil desired level of service anymore. Classical solution to this problem is to build new road infrastructure (urban and intercity highways, bypass roads and new segments of urban roads). That only increases the traffic demand following the saying: “If you built it, they will come!” Additional problem is that in most cases there is no more space available for infrastructure build up. Daily significant congestions appear, mostly in dense populated urban areas. To cope with the heavy congestions, new traffic control approaches are used. These approaches are solutions (services) from the domain of Intelligent Transport Systems, such as ramp metering, variable speed limit control, adaptive control of traffic lights, optimal control for consecutive intersections, etc. In this paper the control of on-ramp traffic flow on urban highways known as ramp metering is examined. To cope with significant daily changes of traffic demand various approaches with autonomic properties like self-learning are applied for ramp metering. One of the approaches with this property is reinforced learning. In this paper the Q-learning algorithm is applied to learn a local ramp metering control law in a simulation environment implemented in the microscopic simulator VISSIM. Proposed approach is tested in simulations with emphasis on a real world situation (typical part of a working day) containing significant changes in traffic demand.

Ramp metering ; Reinforcement learning ; Q-learning ; VISSIM

Rad je objavljen u sklopu konferencije za mlade istraživače u sklopu COST akcije TU1102 "Autonomic Road Transport Support Systems" samo kao zbornik javno dostupan na Internetu.