基于XGBoost的冷水机组不平衡数据故障诊断

FAULT DIAGNOSIS OF UNBALANCED DATA OF CHILLERS BASED ON XGBOOST

冷水机组运行数据分布有不平衡、非高斯、非线性、含噪声的特点,这给基于数据的冷水机组故障诊断带来了挑战。针对这些特点,提出了一种将局部密度过采样算法(Minority Oversampling under Local Area Density,MOLAD)和极限梯度提升算法(e Xtreme Gradient Boosting,XGBoost)相结合的复合算法应用于冷水机组故障诊断中,以克服样本分布不平衡问题,引入代价敏感学习理论来提升重要故障的召回率。基于离心式冷水机组常见的七个故障监测数据进行仿真,结果表明:XGBoost相比于对照组能够更好的对冷水机组状态监测数据进行分类;MOLAD-XGBoost复合模型能够有效处理数据不平衡问题;代价敏感权重可以有效提高重要故障的召回率。

The chiller operating data has unbalanced,non-Gaussian,non-linear,noise-containing characteristics,which poses a challenge for data-based chiller fault diagnosis.Aiming at these characteristics,a chiller fault diagnosis method based on Minority Oversampling under Local Area Density and e Xtreme Gradient Boosting is proposed to chiller fault diagnosis to overcome sample distribution imbalance.Introduce cost-sensitive learning theory to increase the recall rate of important faults.The simulations of seven fault monitoring data commonly used in centrifugal chillers show that XGBoost can better classify chiller status monitoring data compared to the control group.The MOLAD-XGBoost composite model can effectively deal with data imbalance problems;Cost sensitive weights can effectively increase the recall rate for critical failures.