基金项目:上海市“科技创新行动计划”地方院校能力建设专项项目(19020500700);中国华能集团有限公司2019年度科技项目(K-522019007)收稿日期:2021-03-12修回日期:2021-05-21第40卷第2期计算机仿真2023年2月文章编号:1006-9348(2023)02-0108-05基于XGBoost和自适应阈值的电厂风机故障预警夏文苗,黄伟(上海电力大学自动化工程学院,上海200090)摘要:为了对电厂风机实现故障预警,提出了基于极端梯度提升(XGBoost)算法的数据驱动的故障预警方法。首先,通过对电厂原始数据进行数据特征提取和Box-Cox变换,建立基于XGBoost算法的风机轴承温度预测模型;其次,将模型预测值和真实值的偏差用相似度函数表示,并设计了基于区间估计思想的自适应阈值方法;最后利用某电厂送风机数据进行仿真,并将XGBoost算法与支持向量机(SVM)算法、梯度提升树(GBDT)算法进行对比。结果表明该方法能实现风机早期故障预警,验证了该故障预警模型的有效性。关键词:电厂风机;极端梯度提升算法;自适应阈值;相似性;故障预警中图分类号:TM743文献标识码:BPowerPlantFanWindTurbineFaultEarlyWarninginPowerPlantBasedonXGBoostandAdaptiveThresholdXIAWen-miao,HUANGWei(SchoolofAutomationEngineering,ShanghaiUniversityofElectricPower,Shanghai200090,China)ABSTRACT:Inordertorealizefaultearlywarningforwindturbineinpowerplantfan,adata-drivenfaultearlywarningmethodbasedonextremegradientboost(XGBoost)algorithmisproposed.Firstly,throughthedatafeatureextractionandboxCoxtransformationoftheoriginaldataofthepowerplant,thefanbearingtemperaturemodelbasedonXGBoostalgorithmiwasestablished;secondlySecondly,thedeviationbetweenthepredictedvalueandtherealvalueofthemodeliswasexpressedbythroughsimilarityfunction,andtheadaptivethresholdmethodbasedoninter-valestimationideaiswasdesigned;finallyFinally,thedataofapowerplantbloweriswereusedforsimulation,andtheXGBoostalgorithmiswascomparedwiththesupportvectormachine(SVM)algorithmandthegradientliftingtree(GBDT)algorithm.Theresultsshowthatthemethodcanrealizetheearlyfaultwarning,andalsoverifytheef-fectivenessofthefaultwarningmodel.KEYWORDS:Powerplantfan;Extremegradientliftingalgorithm;Adaptivethreshold;Similarity;Faultearlywarning1引言电...