A memetic algorithm for real world multi-intersection traffic signal optimisation problems

Traffic signals play a significant role in the urban transportation system. They control the movement of traffic on urban streets by determining the appropriate signal timing settings. Due to the stochastic nature of the traffic flow, deciding on the best signal timing settings is a computationally complex problem, with the result that traditional analytical methods have been found to be inadequate in dealing with real world scenarios. This issue has already been tackled using computational intelligence algorithms such as the genetic algorithm (GA). However, despite good results, GA may experience slow convergence, especially when dealing with constrained optimisation problems. To address this issue, we propose an adaptive memetic algorithm (MA) for optimising signal timings in real world urban road networks using traffic volumes derived from induction loop detectors. The proposed MA combines the strengths of GA with the exploitation power of a local search algorithm, in an adaptive manner, so as to accelerate the search process and generate high quality solutions. In this work, we propose two important techniques for improving the performance of a traditional MA. First, we use a systematic neighbourhood based simple descent algorithm as a local search to effectively exploit the search space around GA solutions. Second, to achieve a proper balance between the exploration of GA and the local search algorithm, we propose an indicator scheme to control the local search application based on the diversity and the quality of the search process. The proposed MA was tested in two different case studies for the cities of Brisbane, Australia, and Plock, Poland, using the well-known microscopic traffic simulator, AIMSUN. Results demonstrate that our MA is better than GA and traditional fixed-time traffic signal settings.