Dynamic estimation of freeway demand patterns and a stochastic programming approach to freeway ramp metering

Controlling the rate at which vehicles are allowed to enter a freeway from the freeway's on-ramps has become a standard method for reducing the impact of both recurring and nonrecurring traffic congestion. However, existing metering algorithms tend to treat the traffic flow as a deterministic process and to treat the demand for travel on the freeway as fixed and known. In practice however, the demand for travel on a freeway is more accurately modelled as the realization of a stochastic process and there will be uncertainty in the knowledge of the parameters governing this process. This paper concerns two issues: (1) How can data routinely collected by freeway surveillance and control systems be used to estimate the volume and distribution of travellers' demand for freeway travel, and particularly how can the uncertainty concerning this demand be quantified? (2) Given this uncertainty, how ought optimal ramp-metering rates be computed? The first issue is treated as a problem in online systems identification, and both recursive and nonrecursive approaches to the estimation of freeway demand are treated. The second issue is treated as a problem in stochastic programming, in which the demand parameters are treated as random variables. Particular attention is given to dual control issues, in which variable metering rates might be used to reduce uncertaintly concerning the demand parameters.