A neuro-fuzzy algorithm for coordinated traffic responsive ramp metering

Abstract

This paper proposes a nonlinear approach for designing traffic responsive and coordinated ramp control using a self adapting fuzzy system. An adaptive neuro-fuzzy inference system (ANFIS) is used to incorporate a hybrid learning procedure into the control system. The traffic responsive metering rate is determined in every minute by the neuro-fuzzy control algorithm. Coordination between multiple on-ramps is ensured by the integration of a common input into all ramp controllers upstream of a bottleneck and a periodical update of the fuzzy control system in every 15 min. by a hybrid learning procedure. The objective of the online tuning process of the fuzzy parameters is to minimize the total time spent in the system. Therefore, Payne's traffic flow model and a deterministic queuing model are integrated into the control architecture To assess the impacts of the neuro-fuzzy ramp metering algorithm a section of 25 km of the A9 Autobahn was simulated with the FREQ model and compared with two other control scenarios. The results of the simulation of the neuro-fuzzy algorithm are very promising and an implementation of the neuro-fuzzy ramp metering system on a Munich middle ring road within the MOBINET project is planned.