首页|基于注意力机制与XBOA-Bi-LSTM的离心式压缩机故障预警方法

基于注意力机制与XBOA-Bi-LSTM的离心式压缩机故障预警方法

扫码查看
由于离心式压缩机存在着运行工况复杂、维修成本昂贵和长输管道工作环境恶劣的问题,为此,提出了一种基于注意力机制(AM)和蝴蝶算法优化双向长短期记忆神经网络(XBOA-Bi-LSTM)的离心式压缩机故障预警方法.首先,针对传统蝴蝶算法的收敛速度慢、转换概率单一和容易陷入局部最优等问题,通过引入无限折叠迭代混叠映射以丰富蝴蝶算法的初始种群;同时,提出了一种基于种群离散度与迭代次数的自适应惯性转换概率,以提高蝴蝶算法的寻优能力;然后,采用了灰色关联度分析法对测点数据进行了特征提取,结合注意力机制对输入序列进行了灰色关联度系数赋权;最后,建立了双向长短期记忆神经网络故障预警模型,采用仿真实验完成了对离心式压缩机的故障预警;以某天然气长输管道机组的离心式压缩机作为仿真对象,对该离心式压缩机故障预警方法的可行性进行了验证.研究结果表明:采用基于注意力机制与XBOA-Bi-LSTM的离心式压缩机故障预警方法时,在离心式压缩机故障发生前2 h~3h内就发出预警信号,实现了对于离心式压缩机进气过滤器压差异常与支撑轴承工作异常的故障预警目的.
Centrifugal compressor fault warning method based on attention mechanism and XBOA-Bi-LSTM
Aiming at the complex operating conditions,high maintenance costs,and harsh working environment of long-distance pipelines faced by centrifugal compressors,a fault warning method for centrifugal compressors based on attention mechanism(AM)and butterfly algorithm optimized bidirectional short-term and short-term memory neural network(XBOA-Bi-LSTM)was proposed.Firstly,in response to the problems of slow convergence speed,single conversion probability,and easy falling into local optima in traditional butterfly algorithms,infinite folding iterative aliasing mapping was introduced to enrich the initial population of butterfly algorithms.At the same time,an adaptive inertia conversion probability based on population dispersion and iteration number was proposed to improve the optimization ability of the butterfly algorithm.Then,the grey correlation analysis method was used to extract features from the measurement point data,and the grey correlation coefficient was assigned to the input sequence using attention mechanism.Finally,a bidirectional long-term and short-term memory neural network fault warning model was established,and the fault warning of centrifugal compressors was completed through simulation experiments.The feasibility of the fault warning method for a centrifugal compressor in a long-distance natural gas pipeline unit was verified using the centrifugal compressor as the simulation object.The experiment results prove that when using the attention mechanism and XBOA-Bi-LSTM based centrifugal compressor fault warning method,a warning signal is issued within 2 h~3 h before the centrifugal compressor fault occurred,achieving fault warning for abnormal pressure difference of the inlet filter and abnormal operation of the support bearing of the centrifugal compressor.

centrifugal compressorsbutterfly optimization algorithm(BOA)grey relation analysis(GRA)attention mechanism(AM)bidirectional long short-term memory neural network(Bi-LSTM)fault feature extraction

袁镇华、茅大钧、李玉珍

展开 >

上海电力大学 自动化工程学院,上海 200090

上海长庚信息技术股份有限公司,上海 201209

离心式压缩机 蝴蝶优化算法 灰色关联度分析法 注意力机制 双向长短期记忆神经网络 故障特征提取

上海市科技创新行动计划地方院校能力建设专项

19020500700

2024

机电工程
浙江大学 浙江省机电集团有限公司

机电工程

CSTPCD北大核心
影响因子:0.785
ISSN:1001-4551
年,卷(期):2024.41(3)
  • 19