基于传感器技术和I-LSTM算法的风电机设备运行故障检测及诊断研究  

Wind Turbine Equipment Operation Fault Detection and Diagnosis Research Based on Sensor Technology and I-LSTM Algorithm

作　　者：孙晔 郭琳 SUN Ye;GUO Lin(No.9 Port Branch of Qinhuangdao Port Co.,Ltd.Qinhuangdao 066000,China;No.6 Port Branch of Qinhuangdao Port Co.,Ltd.Qinhuangdao 066000,China)

机构地区：[1]秦皇岛港股份有限公司第九港务分公司,河北秦皇岛066000 [2]秦皇岛港股份有限公司第六港务分公司,河北秦皇岛066000

出　　处：《计算机测量与控制》2024年第9期51-57,共7页Computer Measurement &Control

基　　金：河北省科技计划项目(2015ZC20809)。

摘　　要：有效的故障检测与诊断将极大地提高风电机设备运行效率和可靠性,降低维修成本,保障生产过程的顺利进行;为实现高效率的设备故障预警与维护,研究基于传感器技术和机器学习的设备运行故障检测及诊断方法;采用箱型图法和小波包降噪法等对传感器传输的数据信号进行预处理;使用双向长短时记忆网络构建时间序列预测模型;并基于预测残差和贝叶斯概率理论,设计信号异常识别策略,对故障进行实时监测与故障预警;经实验测试,研究设计模型的诊断准确率为98.88%,无漏诊情况,误诊率在1.5%以下,实现了在提前14小时以上进行预警;经实际应用,研究设计模型满足了风电机设备故障预警的及时需求,同时能够在较高的准确率下对故障进行诊断。Effective fault detection and diagnosis will greatly improve the operational efficiency and reliability of wind turbine equipment,reduce maintenance costs,and ensure the smooth progress of production process.To achieve efficient equipment fault warning and maintenance,an equipment operation fault detection and diagnosis method based on sensor technology and machine learning is researched.the box plots and wavelet packet denoising methods are used to preprocess the data signals transmitted by sensors.the bidirectional long short-term memory network is used to construct the time series prediction model.Based on prediction residuals and Bayesian probability theory,a signal anomaly recognition strategy is designed to monitor and warn faults in real-time.Through experimental testing,the diagnostic accuracy of the research and design model is 98.88%,with no missed diagnosis and a misdiagnosis rate of below 1.5%,achieving early warning more than 14 hours in advance.Through practical application,the research and design model meets the timely needs of wind turbine equipment fault warning,and can diagnose faults with high accuracy.

关 键 词：传感器 机器学习 机械设备 故障检测 时序预测 

分 类 号：TP17[自动化与计算机技术—控制理论与控制工程]