基于时移多尺度波动散布熵和改进核极限学习机的水电机组故障诊断  

Fault Diagnosis Method of Hydropower Units Based on Time-shifted Multiscale Fluctuation Dispersion Entropy and Improved Kernel Extreme Learning Machine

作　　者：徐哲熙 刘婷 任晟民 陈建林 吴凤娇[1] 王斌[1] XU Zhexi;LIU Ting;REN Shengmin;CHEN Jianlin;WU Fengjiao;WANG Bin(School of Water Conservancy and Civil Eng.,Northwest A&F Univ.,Yangling 712100,China)

机构地区：[1]西北农林科技大学水利与建筑工程学院,陕西杨凌712100

出　　处：《工程科学与技术》2024年第3期41-51,共11页Advanced Engineering Sciences

基　　金：国家自然科学基金项目(51509210);陕西省重点研发计划项目(2021NY-181)。

摘　　要：水电在能源供给结构改革中承担重要角色,随着风、光、潮汐等新型能源的不断接入,水电机组的负荷运行范围不断加宽,导致水电机组发生事故的概率增加,因此,开展水电机组智能故障诊断研究具有十分重要的现实意义。本文针对水电机组振动信号中蕴含大量噪声信号,干扰故障诊断的问题,提出一种时移多尺度波动散布熵和改进核极限学习机相结合的水电机组故障诊断方法。首先,结合信息熵理论与时移思想,在多尺度波动散布熵的基础上,采用时移理论替代多尺度波动散布熵(MFDE)中传统的粗粒化过程,提出时移多尺度波动散布熵(TSMFDE),通过仿真实验,证明所提方法具有良好的时序长度鲁棒性、抗噪性及特征提取能力,解决了传统多尺度熵粗粒化不足的问题。然后,利用具有可移植性强、寻优能力强和收敛速度快等特征的算术优化算法(AOA)对核极限学习机(KELM)的正则化参数和核函数参数进行寻优,建立AOA-KELM分类器,解决了KELM超参数难以调节的问题。最终,通过转子试验台模拟实验,将TSMFDE提取的特征输入分类器中,完成模式识别工作。仿真结果表明,所提模型取得最高的诊断精度,达到了100.0%,相对于其他流行模型,本文所提模型展现了明显的优势,验证了所提模型的良好诊断精度。Hydropower plays an important role in the reform of energy supply structure.With the continuous access of new energy sources such as wind,light and tide,the load operation range of hydropower units is widening,which leads to the increase of the probability of accidents of hy-dropower units.Therefore,it is of great practical significance to carry out the research on intelligent fault diagnosis of hydropower units.In this paper,aiming at the problem that the vibration signal of hydropower unit contains a large number of noise signals and interferes with fault dia-gnosis,a fault diagnosis method of hydropower unit based on time-shifted multi-scale fluctuation dispersion entropy and improved kernel ex-treme learning machine is proposed.Firstly,based on the multi-scale fluctuation dispersion entropy,the time-shift theory is used to replace the tra-ditional coarse-grained process in the multiscale fluctuation dispersion entropy(MFDE),and the time-shift multiscale fluctuation dispersion en-tropy(TSMFDE)is proposed by combining the information entropy theory and the time-shift idea.Through simulation experiments,it is proved that the proposed method has good timing length robustness,noise resistance and feature extraction ability,and overcomes the problem of insuffi-cient coarse-grained of traditional multi-scale entropy.Then,the arithmetic optimization algorithm(AOA)with strong portability,strong optimiz-ation ability and fast convergence speed is used to optimize the regularization parameters and kernel function parameters of kernel based extreme learning machine(KELM),and the AOA-KELM classifier is established to overcome the problem that the hyperparameters of KELM are diffi-cult to adjust.Finally,through the simulation experiment of the rotor test bench,the features extracted by TSMFDE are input into the classifier to complete the pattern recognition work.The simulation results show that the proposed model achieves the highest diagnostic accuracy,reaching 100.0%.Compared with other popular models,the proposed model

关 键 词：时移多尺度波动散布熵 核极限学习机 算术优化算法 水电机组 故障诊断 

分 类 号：TB126[理学—工程力学]