基于GMM-MD组合算法的过程工业故障预测模型

Fault Prediction Model for Process Industry based on GMM-MD Combinational Algorithm

中文关键词:  故障预测  机器学习  造纸  建模模拟

Key Words:fault prediction  machine learning  papermaking  modeling and simulation

中文摘要:

      介绍了一种基于高斯混合模型（GMM）和马氏距离（MD）组合算法的过程工业故障预测模型。该模型首先通过相关系数去除冗余变量和无关变量，然后通过K-Means聚类算法标记故障前的异常数据以获得核心特征变量，最后基于GMM-MD组合算法构建健康指标，以评估生产过程的健康程度。利用国内某造纸厂实时生产数据对该模型进行验证；结果表明，该模型的故障预测精准率为76.82%，召回率为72.50%，可较好地跟踪造纸过程设备运行状态的变化过程，起到过程工业故障预测作用。

Abstract:

      A process industry fault prediction model based on Gaussian mixture model （GMM） and Mahalanobis distance （MD） combinational algorithm was introduced. The model first removes redundant and irrelevant variables through the correlation coefficient， and then marks abnormal data before the fault through the K-means clustering algorithm to obtain core characteristic variables， and finally constructs health index based on the GMM-MD combinational algorithm to evaluate health degree of the production process. The model was verified by using the real-time production data of a domestic paper mill. The result shows that the predictive accuracy and recall rate of the model is 76.82% and 72.50%， respectively， indicating it could properly track the variation process of equipment running state during papermaking process and play the role of fault prediction in process industry.