针对船舶汽轮机组变负荷过程故障诊断中的耦合参数时序特征难以捕捉以及正常参数变动的干扰等问题，引入TREE-LSTM神经网络模型以实现复杂非线性系统动态数据分类。首先建立某船舶汽轮机组仿真模型，分析并进行故障仿真；随后进行数据预处理与特征工程；最后训练TREE-LSTM模型进行故障诊断，并与SVM、LSTM等模型进行比较。TREE-LSTM模型对于船舶汽轮机组变负荷过程的故障诊断正确率为98.7%，正确率最高。由于引入时间序列与复杂神经网络拓扑结构，TREE-LSTM在处理非线性系统动态数据分类问题时效果更好。
In response to the difficulties in capturing the coupling parameter time series characteristics and the interference of normal parameter changes in fault diagnosis during the variable load process of marine steam turbine units, the TREE-LSTM neural network model is introduced to achieve dynamic data classification of complex nonlinear systems. Firstly, establish a simulation model for a certain marine steam turbine unit, analyze the fault mechanism, and conduct fault simulation; subsequently, perform data preprocessing and feature engineering; finally, a TREE-LSTM model was built for training and fault diagnosis, and compared with models such as SVM and LSTM. The TREE-LSTM model has a fault diagnosis accuracy of 98.7% for the variable load process of marine steam turbine units, with the highest accuracy. It is ultimately believed that due to the introduction of time series and complex neural network topology, TREE-LSTM performs better in dealing with dynamic data classification problems in nonlinear systems.
2024,46(17): 110-115 收稿日期：2023-10-27
DOI：10.3404/j.issn.1672-7649.2024.17.018
分类号：TK269
作者简介：王灏桐(1998-),男,博士研究生,研究方向为基于机器学习的核动力系统、汽轮机组的预测与健康管理
参考文献：
[1] SZYMON G , ANDRZEJ A . Application of a two-stage steam jet injector unit for latent heat recovery of a marine steam turbine propulsion plant[J]. Applied Sciences, 2021, 11(12): 5511–5511.
[2] 张磊, 叶嘉, 林安. 数据机理融合的高精度汽轮机变工况计算模型研究[J]. 舰船科学技术, 2023, 45(3): 96–100.
ZHANG Lei, YE Jia, LIN An. Research on high-precision turbine variable operating condition calculation model based on data mechanism fusion [J]. Ship Science and Technology, 2023, 45(3): 96–100.
[3] WANG J T, WANG S S, PEI B Y, et al. Study on water erosion simulation of low pressure last stage blade of nuclear steam turbine[J]. IOP Conference Series: Materials Science and Engineering, 2019, 504(1): 012057.
[4] 杨楠. 汽轮机智能诊断与健康管理关键技术研究[D]. 北京: 华北电力大学, 2020.
[5] DING X, LIU K, SHI S, et al. Risk assessment of PPP model based on fault tree model for urban underground utility tunnel[C]//Science and Engineering Research Center, Hong Kong. Proceedings of 2019 International Conference on Civil Engineering, Mechanics and Materials Science (CEMMS 2019). DEStech Publications, 2019: 148–160.
[6] FU S , AVDELIDIS P N . Prognostic and health management of critical aircraft systems and components: an overview[J]. Sensors, 2023, 23(19).
[7] JUNRU G. Research on fault diagnosis method of steam turbine generator unit based on multilayer information fusion[C]//IEEE, IEEE Beijing Section. Proceedings of 2013 IEEE 4th International Conference on Software Engineering and Service Science. Institute of Electrical and Electronics Engineers, 2013: 1048–1050.
[8] 范世望, 许伟明, 张祎, 等. 基于LSTM的汽轮机异常状态预警[J]. 电气应用, 2023, 42(3): 56-62.
FAN Shiwang, XU Weiming, ZHANG Yi, et al. LSTM based abnormal state warning for steam turbines[J]. Electrical application, 2023, 42(3): 56-62.
[9] 李晓波, 贾斌, 焦晓峰. 基于高斯混合聚类模型的汽轮机运行状态预警[J]. 内蒙古电力技术, 2020, 38(4): 64-68.
LI Xiaobo, JIA Bin, JIAO Xiaofeng. Steam turbine operation status warning based on Gaussian mixture clustering model[J]. Inner Mongolia Electric Power Technology, 2020, 38(4): 64-68.
[10] WU G, PENG X, YU Z, et al. Determination of the optimum initial operation pressure of a steam turbine unit based on a SVM and GA[C]//International Research Association of Information and Computer Science (IRAICS). Proceedings of the International Conference on Industrial, Mechanical and Manufacturing Science (ICMMS 2014). CRC Press/Balkema, 2014: 124-127.
[11] KUN Y, YING W, SHUANGSHUANG F, et al. Fault mechanisms and diagnosis methods for typical load mutation problems of turbo-generator sets[J]. Frontiers in Energy Research, 2022, 10.
[12] YU W, KIM Y I, MECHEFSKE C. Analysis of different RNN autoencoder variants for time series classification and machine prognostics[J]. Mechanical Systems and Signal Processing, 2021, 149: 107322-107322.
[13] FEI S, YONG L, WEI C, et al. An improved dynamic programming tracking-before-detection algorithm based on LSTM network value function[J]. Systems Science Control Engineering, 2023, 11(1).
[14] WEI L, YONGSHENG L, AKHIL G, et al. Enhancing real-time degradation prediction of lithium-ion battery: A digital twin framework with CNN-LSTM-attention model[J]. Energy, 2024, 286.
[15] LIU W, LIU P, YANG Y, et al. An Attention-Based Syntax-Tree and Tree-LSTM Model for Sentence Summarization[J]. International Journal of Performability Engineering, 2017, 13(5): 775–782.
[16] LI Wanjun, LI Xiaoxia. Analysis and treatment of shutdown due to high condenser pressure towards ultra-supercritical 660MW turbine[J]. IOP Conference Series: Earth and Environmental Science, 2019, 300(4).
[17] JIN H K, WON J B, KYUNGYONG C. Associative knowledge graph using fuzzy clustering and min-max normalization in video contents[J]. IEEE ACCESS, 2021, 9.
[18] W Haotong, L Yanjun, Z Xiaopeng, et al. Research on anomaly detection and positioning of marine nuclear power steam turbine unit based on isolated forest[J]. Nuclear Engineering and Design, 2023, 412.
