基于LSTM常模式阵列的燃气轮机高温部件异常检测方法  

Research on Anomaly detection method of gas turbine hot components based on LSTM pattern array

作　　者：赵帅 罗京 卿顾 吴伟秋 李鑫 龙振华[3] ZHAO Shuai;LUO Jing;QING Gu;WU Wei-qiu;LI Xin;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,四川德阳618000 [2]东方汽轮机有限公司,四川德阳618000 [3]哈尔滨工业大学能源科学与工程学院,黑龙江哈尔滨150001

出　　处：《节能技术》2023年第4期379-384,共6页Energy Conservation Technology

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。研究一种异常检测方法,在高温部件故障早期及时发现部件异常并进行检修意义重大。本文介绍了一种基于LSTM常模式阵列异常检测方法,利用常模式思想提取出反映燃气轮机高温部件性能状态的特征映射,通过LSTM特有的处理时间序列数据的能力,建立LSTM常模式阵列异常检测方法实现高温部件故障早期异常状态灵敏识别。在实际数据上进行方法有效性验证,并与西门子的基于排温保护逻辑的异常检测方法对比,证明了基于LSTM常模式阵列异常检测方法的灵敏性和有效性。The failure of hot components in gas turbines is characterized by a high occurrence rate,strong concealment,and significant destructive impact.Moreover,the maintenance cost and difficulty are substantial once failures occur.Therefore,it is of great significance to study an abnormal detection method that can timely identify component anomalies in the early stage of hot component failure and facilitate repair.This paper presents a novel anomaly detection method based on long short-term memory networks(LSTM)pattern array.By employing the concept of pattern array,it extracts feature mappings that reflect the performance status of hot components in gas turbines.Leveraging the LSTM's capability to process time series data,the LSTM pattern array anomaly detection method is established to achieve sensitive identification of abnormal states in the early stage of hot component failure.The effectiveness of the method is verified using actual data and is compared with Siemens'abnormal detection method based on temperature protection logic,demonstrating the sensitivity and effectiveness of the LSTM pattern array anomaly detection method.

关 键 词：高温部件 异常检测 常模式阵列 长短期记忆网络 

分 类 号：TK471[动力工程及工程热物理—动力机械及工程]