Dynamic Route Guidance Using Neurodynamic Programming

This paper presents a dynamic route guidance method using neurodynamic programming based on the data of GPS-equipped taxies (probe vehicles). Specifically, approximate Q-learning is adopted and ineffective data were eliminated by observing the status of taximeters. The combination of the longitudes and latitudes of the intersections and the time were taken as the states of the algorithm. And the vehicles´ running between the intersections was considered as the transition actions between the states. The neural network was adopted for approximating Q-factors, which denoted the travel time. By observing the probe vehicles´ running in the road network, the neural network was trained so that the approximately optimal route-choice police could be obtained based on it. Finally, simulation on the basis of the electronic map data of Guangzhou city proves that the method is effective on the complex traffic conditions, even as the links´ weights of the road network are unavailable and the state space is continuous