摘要
针对电厂锅炉给水泵因设备老化、运行环境复杂等因素导致振动信号噪声高、特征模糊,进而使现有快速辨识方法精度低、实时性差的问题,提出一种更高性能的振动异常信号智能辨识方法。该方法首先通过滤波与模态分解技术对原始振动信号进行预处理,有效分离并抑制噪声;进而利用随机森林算法对提取的时频域多维特征进行重要性评估与筛选,构建出能够表征异常状态的关键特征子集;最后,通过计算关键特征的熵值并与阈值比较,实现异常信号的快速判定。为验证所提方法性能,在某600 MW智慧电厂进行了实地测试,采集了为期24 h的给水泵振动数据。与现有2种方法对比,实验结果显示,所提方法在10组数据集上的平均误检率低于1%(最低0.22%),且平均辨识时间仅需4.8 ms,在辨识精度与速度上均显著优于对比方法。所提基于随机森林的方法能够快速、准确地辨识锅炉给水泵的振动异常,为智慧电厂实现设备预测性维护、保障安全稳定运行提供了有效的技术手段。
Addressing the issues of high vibration signal noise and ambiguous features in boiler feedwater pumps in power plants,which are caused by factors such as equipment aging and complex operating environments,leading to low accuracy and poor real-time performance of existing rapid identification methods,this paper aims to propose a higher-performance intelligent identification method for vibration abnormal signals.The method first preprocesses the original vibration signals through filtering and modal decomposition techniques to effectively separate and suppress noise.Then,it utilizes the random forest algorithm to evaluate and filter the importance of the extracted multidimensional features in the time-frequency domain,constructing a key feature subset that can characterize abnormal states.Finally,by calculating the entropy value of the key features and comparing it with a threshold,rapid determination of abnormal signals is achieved.To verify the performance of the method,field tests were conducted in a 600 MW smart power plant,collecting vibration data from the feedwater pump for a period of 24 hours.The experiments compared the proposed method with two existing methods.The results showed that the average false detection rate of the proposed method on ten datasets was less than 1%(with a minimum of 0.22%),and the average identification time was only 4.8 milliseconds,significantly outperforming the comparative methods in both identification accuracy and speed.The conclusion indicates that the proposed random forest-based method can quickly and accurately identify vibration abnormalities in boiler feedwater pumps,providing an effective technical means for smart power plants to achieve predictive maintenance of equipment and ensure safe and stable operation.
作者
李仁刚
朱洪伟
王立勇
黄永丰
LI Rengang;ZHU Hongwei;WANG Liyong;HUANG Yongfeng(State Grid Energy Xinjiang Zhundong Coal and Power Co.,Ltd.,Changji 831800,Xinjiang,China)
出处
《自动化技术与应用》
2026年第4期83-86,115,共5页
Techniques of Automation and Applications
基金
新疆维吾尔自治区自然科学基金资助项目(2021D02B012)。
关键词
随机森林
智慧电厂
锅炉给水泵
振动异常
特征选择
信号辨识
random forest
smart power plant
boiler feed pump
vibration anomaly
feature selection
signal identification
