基于FA-LSTM的燃气轮机燃烧室故障演化趋势预估  

Prediction of Gas Turbine Combustion Chamber Failure Evolution TrendBased on FA-LSTM

作　　者：卿硕 袁国凯[1,2] 吴伟秋 傅颖 林朝靖 周之豪 龙振华 QING Shuo;YUAN Guo-kai;WU Wei-qiu;FU Ying;LIN Chao-jing;ZHOU Zhi-hao;LONG Zhen-hua(State Key Laboratory of Clean and Efficient Turbomachinery Power Equipment,Deyang 618000,China;Dongfang Turbine Co.,Ltd.,Deyang 618000,China;School of Energy Science and Engineering,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]清洁高效透平动力装备全国重点实验室,德阳618000 [2]东方汽轮机有限公司,德阳618000 [3]哈尔滨工业大学能源科学与工程学院,哈尔滨150001

出　　处：《汽轮机技术》2024年第5期368-372,378,共6页Turbine Technology

基　　金：清洁透平国家重点实验室共建项目——智能监盘预警的数学模型和大数据处理的预警算法。

摘　　要：燃气轮机高温部件的故障发生率高、隐蔽性强、破坏性大,且故障发生后维修成本高、维修难度大。研究一种故障演化趋势预估方法对于维修人员及时维修和制定维修决策具有重大意义。介绍了一种基于FA-LSTM的故障演化趋势预估方法,结合燃烧室退化机理采用因子分析法(FA)构建健康因子(HI)用于衡量燃机燃烧室的健康状态。利用长短时记忆神经网络(LSTM)特有的处理时间序列数据的能力,预测燃烧室故障演化趋势。以某型燃气轮机为研究对象,并将所提方法与其它6种传统机器学习方法对比,本方法的预测结果MAE和RMSE均最低,能实现准确的退化趋势预测,为短期维护提供有效依据。Gas turbine high-temperature components are prone to failures,which are often concealed and can cause significant damage.Moreover,the repair costs and difficulty are substantial after a failure occurs.Researching a method to predict the evolution trend of failures is of great importance for maintenance personnel to conduct timely repairs and make informed maintenance decisions.This paper introduces a failure evolution trend prediction method based on FA-LSTM,which combines the degradation mechanism of combustion chambers with Factor Analysis(FA)to construct a Health Index(HI)for assessing the health status of gas turbine combustion chambers.Utilizing the Long Short-Term Memory(LSTM)neural network's unique ability to process time-series data,the method predicts the evolution trend of combustion chamber failures.Focusing on a gas turbine as the research subject,this method is compared with six other traditional machine learning methods.The proposed method achieves the lowest Mean Absolute Error(MAE)and Root Mean Square Error(RMSE),enabling accurate degradation trend predictions and providing a reliable basis for short-term maintenance.

关 键 词：燃气轮机 故障演化趋势 因子分析 长短时记忆神经网络 

分 类 号：TK478[动力工程及工程热物理—动力机械及工程] TP183[自动化与计算机技术—控制理论与控制工程]