Keywords: non-deterministic optimisation, traffic lights programming, genetic algorithms, cellular automata, traffic simulation.

Almost any industrialised city in the world shares a common problem: traffic management. The implications of traffic go from the comfort feeling of citizens to the economical development of enterprises. Moreover, the environmental impact of the increasing volume of vehicles cannot be obviated, specially after climate change seems to be proved a reality.

Traffic management is not a single problem but a set of very complex problems, most of them not fully solved so far. One of them is the traffic lights programming optimisation. In [1] traffic lights are proved to be very influencial devices in the traffic behaviour. This is why we chose this matter as research focus.

In [2] we presented a new architecture facing this problem. In a few words, it is composed of three pillars namely a Beowulf cluster - as a MIMD multicomputer backbone - a genetic algorithm - as non-deterministic optimisation technique - and a cellular automata traffic simulator - within the evaluation function of the genetic algorithm.

Through this work our aim is to give an inside view - by means of a visual and statistical study - of what is happening during the evolution of the algorithm until it converges. Recently, we have been provided of real traffic data from the Santa Cruz de Tenerife local government for our research. A wide set of tests have been performed with this data, varying some algorithm parameters.

In this work we present a subset of these test results using 3D pictures like the one shown in Figure 1 for representing the statistical evolution of fitnesses. Every generation a 20 marks histogram was calculated for the fitness values of the whole population. Moreover, a surface to make it easier to view - and to understand - the statistical evolution of the fitness generation by generation was interpolated.

We have presented not only a set of tests tuning the genetic algorithm for a particular np-complete optimisation problem, but also a methodology - such a visual one - helping the researcher tune the genetic algorithm with a much deeper knowledge of how it is done than before.

1

E. Brockfeld, R. Barlovic, A. Schadschneider, M. Schreckenberg, "Optimizing Traffic Lights in a Cellular Automaton Model for City Traffic", http://arxiv.org/ps/cond-mat/0107056, (2001).

2

J. Sanchez, M. Galan, E. Rubio, "Genetic Algorithms and Cellular Automata: A New Architecture for Traffic Light Cycles Optimization", Proceedings of The Congress on Evolutionary Computation 2004 (CEC2004) (2004) Volume II, 1668-1674. doi:10.1109/CEC.2004.1331096

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £105 +P&P)