A dynamic route guidance arithmetic based on reinforcement learning

The links´ weights of road networks are important parameters in the past researches on dynamic traffic route guidance. But they are usually unavailable on the complicated traffic conditions. This research is focused on the development of a dynamic route guidance arithmetic based on reinforcement learning. The arithmetic need not links´ weights and made used of data provided by the probe vehicles equipped with GPS sensors. The numbers of the intersections and the time segments were taken as the states. The vehicles´ running actions between the intersections were considered as the transition actions from one state to another, and the time costs were described by Q-factors. By observing probe vehicles running in road network, Q-factors were iterated so that the optimal route-choice police could be obtained. Finally, the simulation was processed based on the electronic map data of Guangzhou city. The results indicate that the arithmetic is quite effective on the complex traffic conditions.