Analysis of frequency-based compact genetic algorithm (fb-cGA)

作者：Rimcharoen Sunisa; Phiromlap Srichol; Leelathakul Nutthanon^*

来源：Maejo International Journal of Science and Technology, 2015, 9(1): 121-135.

DOI：10.14456/mijst.2015.10

• compact genetic algorithm
• updating strategy
• update frequency
• update continuity

A behaviour analysis of frequency-based compact genetic algorithm (fb-cGA) is proposed. The fb-cGA is a version of compact genetic algorithm (cGA) enhanced by the use of a new updating strategy. The algorithm counts the number of probability updates and the continuities of probability-update directions and uses them to adaptively update the algorithm's step sizes. This method requires fewer function evaluations and achieves solutions that are more accurate than those from the conventional cGA. It has been shown that fb-cGA can reduce the number of function evaluations to only one ninth of the number obtained from cGA on ten copies of a 3-bit trap function using a tournament size of 2. We conduct parameter studies and show that the use of one fourth of the population size (psize/4) as the algorithm's starting threshold can improve the overall efficiency of fb-cGA. The behaviour of fb-cGA on various problems is also examined. The results of the analysis show that information from the algorithm's past experience (i.e. the numbers of probability updates and continuities) can help the fb-cGA to update the probability vector towards a more promising direction, requiring fewer function evaluations.

• 出版日期2015-4