贝叶斯优化算法的选择策略分析

Analysis on selection strategy of Bayesian optimization algorithm

针对贝叶斯优化算法的选择策略问题,对变量无关,双变量相关,多变量相关等3类典型函数分别用锦标赛选择、截断选择和比例选择以及自适应比例选择进行了实验。建立了相应的贝叶斯网络概率模型,并分析指出锦标赛选择策略能有效保持样本的多样性,并能建立起准确的网络模型。与比例选择策略和截断选择策略相比较,该选择策略更适用于贝叶斯优化算法。

To test selection strategy on Bayesian optimization algorithm,tournament selection,truncation selection,proportional selection and self-adaptive proportional selection are adapted to deal with typical dependency-free function,bivariate dependencies function and multivariate dependencies function,corresponding Bayesian network is set up.Analysis discovers that the diversity of samples is kept and the model is more accurate in tournament selection.Tournament selection is the best selection strategy for Bayesian optimization algorithm,truncation selection and proportional selection are unsuitable for the algorithm.