非线性信息融合技术在火力发电机组故障诊断中的应用

Application of Nonlinear Information Fusion Technology in Fault Diagnosis of Thermal Power Unit

火力发电机组的振动信号往往是多种振源信号的非线性混叠,由此给火力发电机组故障诊断中特征信号的提取与分析带来了强烈的干扰。针对此问题,提出了基于小波-非线性独立分量分析(independent componentanalysis,ICA)的火力发电机振动信号非线性盲分离与特征提取新方法。首先,利用小波去噪技术消除加性噪声的影响;然后,通过径向基函数(radial basis function,RBF)神经网络,并结合线性ICA算法估计去噪信号的非线性混合解混函数,实现信号的非线性盲分离,得到火力发电机振动故障的关键信号源;最后,利用小波包分解提取分离信号的统计特征并作为识别损伤状态的有效参数,应用RBF神经网络分类器对火力发电机故障类型进行智能识别。对某发电站的火力发电机实际故障振动信号进行实验分析,结果表明,所提出的非线性盲分离模型能够从含有加性噪声的非线性振动源混合观测信号中提取故障振动源,得到故障信号的可靠特征,取得较好的故障诊断效果,且故障检测精度比线性盲分离技术提高了4.4%以上。

Vibration signal of thermal power unit is often a kind of nonlinear mixture of different vibration source which may cause strong interference for extracting and analyzing characteristic signal in fault diagnosis of thermal power unit. Therefore, this paper proposes nonlinear blind source separation and feature extraction method for vibration signal of thermal power unit based on wavelet-nonlinear independent component analysis. Firstly, it uses wavelet denoising technology to remove additive noise, then by radial basis function neural network and combining linear ICA algorithm to evaluate nonlinear mixed solution of mixed function in order to realize nonlinear blind source separation and acquire key signal source of vibration fault. At last, it uses wavelet packet to decompose and extract statistic feature of separative signal which is taken as effective parameter for distinguishing faulted condition and applies RBF neural network classifier to intelligently distinguish fault types of the thermal power unit. By testing and analyzing practical faulted vibration signal of thermal power unit in one substation, it shows that the proposed nonlinear blind separation model is able to extract faulted vibration source in mixed observed signals of nonlinear vibration source with additive noise, acquire reliable characteristic of the faulted singal and good fault diagnosis effect with improved fault detection precision of 4.4%.