关键词: 属性重要度, 贝叶斯判别分析, 主成分分析, 石油钻井安全评价

Abstract: In the industrial safety evaluation, the hybrid method with principal component analysis and discriminant analysis is a widespread approach. PCA-BDA is an effective method of this type. However, this method without comprehensive results to show the causality between the attributes and the safety level, it is essentially a grey box. Therefore, an improved PCA-BDA method is been proposed that introduces the measure of attribute importance degree. Applied it to the safety evaluation of petroleum drilling to identify the critical attributes, thus to avoid or reduce risks by controlling the critical attributes targeted. First, the safety level is evaluated with the original PCA-BDA. Second, the importance degree of each attribute is computed with the eigenvector matrix in PCA, the classification function coefficients in BDA, and the weight of safety levels. Experiments are undertaken on a real-world petroleum drilling firm data set. Results show that we reduce 70% of the safety level effectively by controlling the critical attributes, while the safety level hardly change by adjusting the least important attributes.

Key words: attribute importance degree, Bayes discriminant analysis (BDA), principle component analysis (PCA), petroleum drilling safety evaluation