Decentralized artificial intelligent traffic control

Abstract

A generic decentralized traffic junction controller using artificial intelligence is constructed. The controller is simple in structure and does not require any geographic knowledge of the junction. Each junction has a decentralized traffic junction controller that can be connected together to form traffic network control. Number of controllers can join together to form a large network. The controller consists of 3 modules, a local decision module, an optimization module and a junction coordination module. The local decision module is a fuzzy controller with a hierarchical structure. The optimization module use Genetic Algorithms to generate fuzzy rules for the local decision module. The junction coordination module adopts real time dynamic programming to obtain coordinated result. Several optimization techniques are used to enhance the performance of the controller, such as hierarchy and time alignment. Cross-junction with turning traffic is modeled. Computer simulation is used to evaluate the performance of the controller against fixed-time approach. The controller can improve up to 20% and 8% for single and multiple junction environments respectively. Other approaches that can further improve the controller are also discussed at the end.