Bill Keller, Rudi Lutz
This paper describes experiments with a new genetic operator, randomised and/or crossover (RAOC), which shows good performance on a range of function optimisation problems. Originally motivated by work on grammatical inference, RAOC was observed to outperform more traditional crossover operators both in terms of its ability to locate global optima in the search space and in terms of its speed (measured as the number of function evaluations taken to reach a solution). This paper describes an extensive empirical study comparing RAOC with a number of more standard operators on a range of function optimisation problems. The results of this study bear out the earlier observations and demonstrate that RAOC is of wider applicability than to the original grammatical inference problem which motivated it.
Download compressed postscript file