1. 中国大唐集团科学技术研究院有限公司华东电力试验研究院
2. 设备状态监测与控制教育部重点实验室(华北电力大学)
3. 河北工程大学水利水电学院
4. 安徽大学物质科学与信息技术研究院
纸质出版:2025
移动端阅览
王远鑫, 陈鸿伟, 杨新, 等. 基于注意力机制CNN-BiLSTM的磨煤机故障预警方法[J]. 华北电力大学学报(自然科学版), 2025,(4):134-142.
磨煤机是火电机组的重要辅机之一,其运行状态对机组的稳定出力和安全运行发挥着重要影响,所以研究故障预警方法可以有效降低磨煤机的故障停机概率,提升机组的整体可靠性。结合多种深度学习方法的优势,提出了基于注意力机制的卷积神经网络-双向长短期记忆神经网络(CNN-BiLSTM)的故障预警方法。通过CNN学习数据的空间特征,通过BiLSTM学习数据的时间特征,最后通过注意力机制为提取到的时空特征根据重要度分配不同的权重,增强重要特征在计算过程中的作用,从而提升模型的预测性能。使用某660 MW火电机组的磨煤机运行数据进行验证,结果表明所提方法在正常阶段相比对比方法具有更高的预测精度,在故障阶段预测残差的上升趋势更加明显。为了进一步提升故障预警效果,引入相对熵指标对残差进行处理,实验结果表明相对熵指标解决了残差控制图的缺点,有效提升了故障预警效果和预警时间。所提方法提前33 min预警了磨煤机轴承故障。
The coal mill is one of the important auxiliary machines in thermal power units
and its operating status plays a significant role in the stable output and safe operation of the unit. Therefore
researching fault warning methods can effectively reduce the probability of unexpected shutdown of the coal mill and improve the overall reliability. By integrating the advantages of various deep learning methods
we propose a fault warning method based on attention mechanism CNN-BiLSTM. CNN is used to learn spatial features
and BiLSTM is used to learn temporal features. Finally
attention mechanism assigns different weights to the extracted spatiotemporal features based on their importance
thereby amplifying the influence of critical features during the computation process and elevating the model's predictive performance. A 660 MW thermal power unit's coal mill is used as a case study. The validation results show that the proposed method exhibits higher prediction accuracy than the comparison methods during the normal stage and a more pronounced upward trend in prediction residual during the fault stage. In order to further improve the performance of fault warning
we introduce the relative entropy index to process the residual. The experimental results show that the relative entropy index solves the shortcomings of the residual control chart and effectively improves the fault warning effectiveness and warning time. Specifically
the proposed method can issue a warning for coal mill bearing faults 33 minutes ahead of time.
袁存波,田亮.基于改进证据理论的中速磨煤机故障诊断[J].华北电力大学学报(自然科学版),2022(06).
杨婷婷,高乾,李浩千,吕游,陈晓峰.基于卷积神经网络-长短时记忆神经网络的磨煤机故障预警[J].热力发电,2022(10).
张烨,黄伟.基于天牛群算法优化SVM的磨煤机故障诊断[J].机电工程,2022(03).
孙立,Fengqi You.机器学习和数据驱动算法在智慧发电系统中的应用——一种不确定性处理的视角[J].Engineering,2021(09).
姜兴华,韩万里,茅大钧.基于核主元分析和GRU神经网络的电站辅机故障预警研究[J].热能动力工程,2021(07).
常永亮.基于多变量状态估计的磨煤机一次风量故障预警仿真[J].山西电力,2021(03).
孙同敏.基于ISAE的磨煤机故障预测与诊断方法研究[J].煤炭工程,2021(05).
王梓齐,张书瑶,刘长良.基于增量式相对熵的风电机组实时状态监测[J].电子测量与仪器学报,2020(12).
游大宁,郝旭东,张国强,程定一,王进.基于热电解耦的电力现货市场机制研究[J].中国电机工程学报,2021(09).
刘长良,张丛,王梓齐.基于条件互信息的NO_x生成量模型变量-时滞联合选择方法[J].华北电力大学学报(自然科学版),2019(06).
李晓彬,牛玉广,葛维春,罗桓桓,周桂平.基于改进堆叠自编码网络的电站辅机故障预警[J].仪器仪表学报,2019(06).
肖黎,罗嘉,欧阳春明.基于半监督学习方法的磨煤机故障预警[J].热力发电,2019(04).
孙栓柱,江叶峰,董顺,周挺,代家元,李益国.基于灰箱模型的中速磨煤机故障诊断方法[J].动力工程学报,2018(03).
李晓彬.低负荷下电站辅机安全监控技术研究与应用[D].华北电力大学(北京),2020(06).
Chang Yuqing,Liu Leyuan,Kang Xiaoyun,Wang Fuli.Decomposition-Based Bayesian Network Structure Learning Algorithm for Abnormity Diagnosis Model for Coal Mill Process[J].Electronics,2022.
Zhang Hui,Pan Cunhua,Wang Yuanxin,Xu Min,Zhou Fu,Yang Xin,Zhu Lou,Zhao Chao,Song Yangfan,Chen Hongwei.Fault Diagnosis of Coal Mill Based on Kernel Extreme Learning Machine with Variational Model Feature Extraction[J].Energies,2022.
Yong Hu,Boyu Ping,Deliang Zeng,Yuguang Niu,Yaokui Gao,Dongming Zhang.Research on fault diagnosis of coal mill system based on the simulated typical fault samples[J].Measurement,2020.
Pengcheng Zhu,Hong Qian,Tingting Chai.Research on early fault warning system of coal mills based on the combination of thermodynamics and data mining[J].Transactions of the Institute of Measurement and Control,2020.
Jian Yifan,Qing Xianguo,Zhao Yang,He Liang,Qi Xiao.Application of Model-Based Deep Learning Algorithm in Fault Diagnosis of Coal Mills[J].Mathematical Problems in Engineering,2020.
0
浏览量
714
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621