Automation of fuzzy systems for intelligent traffic lights

Abstract

Intense vehicle traffic is one of the main disorders in large cities, and in Brazil, where the responsible authorities have not trained the road networks, overcrowding in traffic causes even more obstacles. The applications of computational intelligence techniques in traffic are very broad, with emphasis on smart traffic lights. For the design of intelligent traffic lights, this work proposes the use of Fuzzy Logic, and has as main objective the automatic generation of fuzzy systems using evolutionary fuzzy models for this purpose. To achieve this objective, the traffic simulation software SUMO is used, which allows the elaboration of scenarios of intersections controlled by traffic lights. In these scenarios, the traffic performance is evaluated based on different adjustments in the membership functions and in the set of rules of the fuzzy system that controls the traffic lights, and these adjustments are performed by Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). When comparing the traffic performance with traffic lights controlled by fuzzy and fuzzy with optimized hyperparameters, there are important improvements in the analyzed traffic variables, such as waiting time and car queue size/length, in addition to reducing the emission of toxic gases and fuel consumption. Thus, this work highlights the importance of employing evolutionary fuzzy models in hyperparameters optimization.