小波能量偏度神经网络的HVDC换相失败故障诊断

Fault Diagnosis of Commutation Failures in HVDC System Based on Neural Network of Wavelet Energy Skewness

为了有效诊断引发换相失败的故障原因,提出了一种基于小波能量偏度神经网络的高压直流输电系统换相失败故障诊断新方法。首先对采集到的逆变侧直流电压故障信号进行15层小波分解,获取各尺度下的小波变换系数,提取小波能量偏度;然后构造信号的小波能量偏度特征向量,并以此向量作为故障样本对3层BP神经网络进行训练,实现换相失败故障原因诊断。以某±800kV特高压直流系统为例,通过对引发换相失败故障的多种原因进行仿真分析,用该方法进行小波分解、故障特征提取和BP网络训练,并对某未知故障进行识别。结果表明,该方法能准确诊断出引发换相失败的故障原因。

Commutation failure is a common fault in HVDC systems. In order to effectively diagnose the causes of com- mutation failures, a new fault diagnosis method of commutation failures in the HVDC system is proposed based on the neural network of wavelet energy skewness. After the 15-layer wavelet decomposition of acquired DC voltage fault signal in the inverter side is performed, the wavelet coefficients on every scale are obtained and the wavelet energy skewness is extracted to construct the eigenvector of wavelet energy skewness. With this eigenvector as fault sample, three-layer BP neutral network is trained to implement the fault diagnosis of commutation failures. A ± 800 kV UHVDC system is taken as an example, and a variety of causes inducing failure fault are simulated and analyzed. This method is used to conduct wavelet analysis, fault feature extraction and BP neural network training. At last, the unknown fault is identified. The results show that the method can accurately diagnose the fault cause of commutation failures.