Abstract: Characteristic variables selection is a critical factor that affects the result of discriminant analysis. Appropriate selection of variable combination can improve the correct rate and reduce the amount of computation. This paper introduces the basic principle of Bayes discriminant and stepwise discriminant method, analyses some methods of characteristic variables optimization at present, takes application of Bayes discriminant in oil gas interpretation and evaluation as an example to research on variable optimization method in stepwise Bayes discriminant analysis, and puts forward a variable step optimization strategy and a multi-model optimization strategy, including variable range selection, data preprocessing, feature extraction, preliminary screening of variables and stepwise discriminant, which makes correct judgement rate optimized gradually, and obtains more satisfactory discriminant result.

Key words: Bayes discriminant, stepwise discriminant, variable optimization, discriminant factor, gas logging

摘要： 判别分析中特征变量是影响判别结果的决定性因素，选取适当的特征变量组合可以提高正判率、减少计算量。介绍了贝叶斯判别和逐步判别法的基本原理，分析了目前出现的一些特征变量优化方法，以油气解释评价中的贝叶斯判别应用为例，对于逐步贝叶斯判别中的变量优化方法进行了研究和总结，提出了变量的多步优化策略和分步多模型优化策略，包含了从变量范围选择、数据预处理、特征变量提取到初步筛选和逐步判别的完整过程，使得正判率不断优化，最终得到了较为满意的判别结果。

关键词: 贝叶斯判别, 逐步判别, 变量优化, 判别因子, 气测录井