基于偏鲁棒M-回归的间歇过程离群点检测

以多元统计分析技术为核心的间歇过程建模、在线监测逐渐成为过程工业的关注焦点，然而过程数据中存在的大量离群点将直接影响上述方法的可靠性，为此提出了一种基于偏鲁棒M-回归的间歇过程离群点检测方法.首先基于极大相关熵估计建立鲁棒预测模型；然后利用偏鲁棒M-回归算法计算模型的回归系数；最后采用Hampel识别器分析最终的权值，从而实现离群点的检测.将所提方法应用于某间歇反应过程，实验结果验证了方法的有效性.

Outlier Detection for Batch Processes Based on Partial Robust M Regression

Batch processes modeling， online monitoring with multivariate statistical analysis at the core have gradually became the research focus of process industry， however， the reliability of such methods would be affected by a large number of outliers in process data. Thus， an outlier detection method for batch processes based on partial robust M regression was proposed to solve this issue. First， the robust prediction model was established based on maximum correntropy estimator. Then partial robust M regression algorithm was utilized to calculate the model regression coefficients. Finally， Hampel identifier was used to analyze the final weights， and the outlier detection was fulfilled. The proposed outlier detection method was applied to a batch reaction process， and the experimental results indicated that effectiveness of the proposed method.