Page 97 - 水利学报2021年第52卷第4期

［ 16］ 陈后金 . 数字信号处理［M］. 3 版 . 北京：高等教育出版社，2018 .
［ 17］ 卢娜 . 基于多小波的水电机组振动特征提取及故障诊断方法研究［D］. 武汉：武汉大学，2014 .
［ 18］ MALLAT S . A Wavelet Tour of Signal Processing［M］. Pittsburgh：Academic Press，1998 .
［ 19］ LEI Y，HE Z，ZI Y，et al . Fault diagnosis of rotating machinery based on multiple ANFIS combination with Gas
［J］. Mechanical Systems & Signal Processing，2007，21（5）：2280-2294 .
［ 20］ IGBA J，ALEMZADEH K，DURUGBO C，et al . Analyszing RMS and peak values of vibration signals for condi⁃
tion monitoring of wind turbine gearboxes［J］. Renewable Energy，2016，91（3）：90-106 .
［ 21］ 乔国华，郭路遥，吴一敌，等 . 基于遗传优化最小二乘支持向量机的变电站全寿命周期成本预测模型［J］.
中国电力，2015，48（11）：142-148 .

Research on on-line evaluation method of state degradation
of hydropower unit based on vibration signal

1 1 2 2
LIU Dong ，LAI Xu ，HU Xiao ，XIAO Zhihuai
（1. State Key Laboratory of Water Resources and Hydropower Engineering Science，Wuhan University，Wuhan 430072，China；
2. Key Laboratory of Hydraulic Machinery Transients，Ministry of Education，Wuhan University，Wuhan 430072，China）

Abstract： Realizing the deterioration assessment and the fault early warning of hydropower units is a hot
topic in the industry. In this paper， an online evaluation method for unit degradation combining time do⁃
main and frequency domain features is proposed. （1） The detection index is used to determine the time do⁃
main characteristics of the vibration signal that are most sensitive to the operating state of the unit. The op⁃
erating parameter X （head， opening， etc.） of the unit in health state and the time domain characteristics
Y of the vibration signal selected by the detection index are selected as healthy sample， and the least
X
squares support vector machine is used to construct the unit state health model Y = f ( ) . The real-time
operating condition parameters are inputted into this model，and the health value of the unit's vibration sig⁃
nal time-domain characteristic under corresponding operating conditions is predicted online. The relative er⁃
ror between the health value and the actual value is calculated as a time-domain degradation index. （2）
The wavelet transform and singular value theory are used to decompose the vibration signal to extract the
singular value feature vector of the unit's vibration signal under healthy state and obtain the health cluster⁃
ing center. The relative Euclidean distance is used as an index of frequency domain degradation. Combined
with the time and frequency domain degradation indicators，an online calculation of comprehensive degrada⁃
tion indicators is performed to assess the degree of unit degradation at the current moment. Combined with
actual unit operation cases，the effectiveness and practicability of the model are verified.
Keywords： hydropower unit； degradation assessment； vibration signal； detection index； least squares sup⁃
port vector machine；wavelet singular value

（责任编辑：杨 虹）

— 473 —