遗传算法优化性能评价准则研究

为了克服传统遗传算法优化性能评价准则（如在线性能、离线性能）忽略随机因素对算法的影响，从而不能准确评价算法性能的缺点，提出了一种基于平均偏离距和偏离距标准差的新的遗传算法优化性能评价准则，给出了平均偏离距和偏离距标准差的数学定义，并分析了它们的泛函意义．由于平均偏离距和偏离距标准差采用遗传算法多次运行结果的统计参数来评价算法的性能，因此能够较好地消除随机因素对算法性能的影响．同时，应用所提出的评价准则研究了二进制码和格雷码对遗传算法优化性能的影响．基于F2函数的数值实验结果表明，与二进制码相比，格雷码的平均偏离距和偏离距标准差指标都比较低，因此能够更好地提高遗传算法的优化性能．

Evaluation for Genetic Algorithm Performance

The traditional evaluation criteria for genetic algorithm(GA) performance,such as on-line performance and off-line performance,usually ignore the stochastic search characteristics to seem unable to assess the GA performance accurately.It is necessary to propose a new evaluation criterion,where two indexes,mean error(ME) and standard deviation of error(SDE),are introduced and mathematically defined.As the statistical parameters of GA searching,ME and SDE indicate the GA performance more perfectly.The performances of binary encoding GA and Gray encoding GA are investigated comparatively with the proposed evaluation criterion,and the results show that Gray encoding GA outperforms binary encoding GA.