基于FDM孪生网络的风电场集电线单相接地故障区段定位

Single-phase grounding fault location of wind farm collection line based on FDM and Siamese neural network

针对风电场集电线路故障后定位困难以及故障样本稀缺问题,文章提出了基于傅里叶分解算法与孪生神经网络相结合的风电场集电线路区段定位方案。傅里叶分解算法适用于集电线路接地故障所产生的瞬态非平稳信号;孪生神经网络可用于对样本集的扩充和辅助训练识别网络。该方案首先提取集电线故障信号线模分量,借助傅里叶分解算法生成时频能量谱;然后借助孪生神经网络扩充样本集,并在该集合上辅助训练定位网络;最后将孪生神经网络分支部分保留以形成定位网络,基于已有故障区段模态,应用定位网络完成对未知故障模态的判别。试验结果表明,文章提出的算法在小样本情况下比传统行波法更适合风电场集电线路故障定位。

To solve the difficulties of locating fault points and the scarcity of fault samples after faults of wind farm collector lines,a section location scheme for wind farm collector lines based on the combination of Fourier decomposition algorithm and siamese neural network is presented.The Fourier decomposition algorithm is applicable to the transient non-stationary signal caused by the grounding fault of the collection line.Siamese neural networks can be used to expand the sample set and assist in training recognition networks.The important process of this scheme includes:first,extracting the line mode component of the collector line fault sign,and generating time-frequency energy spectrum with fourier decomposition algorithm.Then,the sample set is expanded with the siamese neural network and the locating network is trained on this set.Finally,the branch of the siamese neural network is reserved to form a locating network.The locating network is used to identify unknown failure modes based on the existing failure segment modes.Experiments show that the proposed algorithm is more suitable for wind farm collector line fault location than traditional traveling wave method in small sample cases.