Dynamic programming based coordinated ramp metering algorithms : a thesis presented in partial fulfillment of the requirements for the degree of PhD of Englineeriing in Mechatronics at Massey University, Auckland, New Zealand

Abstract

Motorway congestion can be classified into two types, recurrent congestion and non-recurrent congestion. Recurrent congestion happens during peak hours. Non-recurrent congestion occurs due to car accidents, weather conditions or public events. Negative impacts of traffic congestion include wasted fuel, pollution, travel delay and spillover effects caused by slow traffic. Ramp metering, as an only way to regulate traffic amount accessing to the motorway, is considered as the most cost-effective way to prevent the recurrent congestion. Coordinated ramp metering was developed to control a number of on-ramps simultaneously to improve traffic conditions on busy motorways. The existing coordinated ramp metering algorithms were normally established on macroscopic traffic flow models based on Payne’s wok, the performances of which were measured by the employed macroscopic model themselves, and the released metering rates of which tended to be continuous. Implementations in microscopic traffic simulators were few. This thesis presents DP (Dynamic Programming) based online control approaches for the optimal coordination of ramp metering and evaluates its performances in both macroscopic and microscopic traffic simulation environment. DP decision networks were proposed, where a traffic system can be modeled as a number of discrete traffic states and separated by time stages, and the control problem of coordinated ramp metering was treated as the minimization problem to search the optimal trajectory of discrete decision variables (ramp metering rates) that minimized a cost criterion in terms of TTS (total time spent) along the time horizon. Experiments conducted in the macroscopic simulation environment demonstrated the full potential of proposed algorithms with precise queue constrains in an ideal deterministic environment, and experiments conducted in the microscopic simulation environment indicated the performances of the proposed algorithms in a stochastic environment and revealed the feasibility in the real world. The implementation of DP ramp metering was proposed under the framework of receding horizon control. A 6.7km stretch of motorway in Auckland, New Zealand, was chosen as a study location and constructed by a microscopic simulator as a simulation scenario and by a macroscopic traffic model as a prediction model. The simulation results indicated that the proposed algorithms were able to eliminate motorway queues under high traffic demands and manage queue lengths at metered on-ramps when queue constrains were not overstrict. The simulation results also revealed that 9 discrete metering rates for each ramp meter were adequate to prevent motorway queues. Such feature not only proved that the optimal trajectory converged very fast in the proposed DP decision networks, but also made on-line control system possible due to less computational load.