应用集成学习方法来解决多元统计过程控制中的质量诊断问题:当HotellingT~2控制图发出报警信号时,判断哪一个变量或者是哪些变量组合导致过程异常的发生。这种集成学习方法采用分类与回归树(Classification and Regression Trees,CART)为基分类器,沿着误判损失函数的梯度下降方向逐步优化基分类器组合,并基于误判损失最小的原则判断异常信号的类别。实例研究表明,这种集成学习方法具有较好的诊断效果,在诊断准确率方面优于已有的方法。 Application of integrated learning method to solve the quality of multivariate statistical process control diagnostic problems: When HotellingT ~ 2 control chart issued a warning signal, determine which variable or which combination of variables lead to the occurrence of abnormal process. This integrated learning method uses the Classification and Regression Trees (CART) -based classifier to optimize the base classifier combinations gradually along the gradient descent direction of the misjudgment loss function and to judge the anomalies based on the principle of minimizing miscarriage of justice The type of signal. The case study shows that this integrated learning method has better diagnostic results and is superior to the existing methods in the accuracy of diagnosis.