Optimising Real-World Traffic Cycle Programs by Using Evolutionary Computation

Traffic congestion, and the consequent loss of time, money, quality of life, and higher pollution, is currently one of the most important problems in cities, and several approaches have been proposed to reduce it. In this paper, we propose a novel formulation of the traffic light scheduling problem in order to alleviate it. This novel formulation of the problem allows more realistic scenarios to be modeled, and as a result, it becomes much harder to solve in comparison to previous formulations. The proposal of more advanced and efficient techniques than those applied in past research is thus required. We propose the application of diversity-based multi-objective optimizers, which have shown to provide promising results when addressing single-objective problems. The wide experimental evaluation performed over a set of real-world instances demonstrates the good performance of our proposed diversity-based multi-objective method to tackle traffic at a large scale, especially in comparison to the best-performing single-objective optimizer previously proposed in the literature. Consequently, in this paper, we provide new state-of-the-art algorithmic schemes to address the traffic light scheduling problem that can deal with a whole city, instead of just a few streets and junctions, with a higher level of detail than the one found in present studies due to our micro-analysis of streets.