【摘要】 为了解决统计逐步判别分析法存在的问题,提出了一种基于粗糙集属性约简的统计判别分析方法.首先采用粗糙集属性约简进行变量筛选,这样可充分利用粗糙集属性约简不需要属性分布的先验信息这一特点,再对所选择的变量进行Bayes判别分析训练,建立判别函数或相应的后验概率函数,以解决选择变量过程中存储量较大且检验变量的重要性总体服从正态分布这一主观性假设等问题.通过对油气储层数据的实际分析表明,所提方法不仅易于实施,而且检验数据集的判别准确率高于统计逐步判别分析法,同时可节省预测成本,提高预测速度.更多还原

【Abstract】 In order to overcome some shortcomings of the statistical stepwise discrimination method,a new discrimination method based on rough set attribute reduction is proposed,in which the attribute reduction of rough set is used to select important discriminating variables by sufficiently taking advantage of the feature that the attribute reduction of rough set needs no prior information on the distribution.Then,with the selected variables,Bayesian discrimination procedure is used to build the discrimination function or compute the posterior probability so as to solve the problems that the storage size is much bigger and the importance of discriminating variables wholly obey the normal distribution from the subjective assumption in the statistical stepwise discrimination method.Through applying the proposed method to the real-world dataset of reservoir of oil and gas,it is demonstrated that the proposed method not only is easy to be implemented with higher correct discrimination rate,but also can decrease the computational cost and increase the prediction speed.更多还原