输电线路发生各种类型故障时将严重威胁系统的安全稳定运行，传统基于故障录波的输电线路故障诊断方法侧重于利用故障录波提供的零序电流信息，难以准确辨识所发生故障的类型。文章提出了基于样本不均衡和特征优选多源融合的(kepler optimization algorithm-convolutional neural network-bidirectional long short-term memory network-selfattention，KOA-CNN-BiLSTM-Selfattention)输电线路故障类型辨识方法，获取输电线路分布式在线监测装置提供的行波数据，整合形成输电线路的实际有效故障波形数据库及带标签信息数据库；基于融合改进(synthetic minority over-sampling，technique，SMOTE)算法进行样本均衡，采用时域函数、快速傅里叶变换(fast fourier transform，FFT)、小波包分析、波形函数提取不同输电线路故障类型的时域、频域、时频域及波形脉宽等故障特征，基于Deep Lasso(deep least absolute shrinkage and selection operator)回归完成特征优选再进行多源信息融合，最后采用KOA-CNN-BiLSTM-Selfattention算法完成输电线路的故障类型辨识。通过基于输电线路历史故障跳闸数据的算例验证了所述方法的正确性和有效性。
英文摘要:
The occurrence of various types of faults in transmission lines will seriously threaten the safe and stable operation of the system, and the traditional fault diagnosis methods for transmission lines based on fault recordings focusing on the use of zero-sequence current information provided by fault recordings, which makes it difficult to accurately identify the types of faults that occur. This paper proposes a transmission line fault type identification method of kepler optimization algorithm-convolutional neural network-bidirectional long short-term memory network-selfattention, (KOA-CNN-BiLSTM-Selfattention) based on multi-source fusion with sample imbalance and feature preference. The travelling wave data provided by the distributed online monitoring device of transmission line is obtained, integrating to form a database of actual effective fault waveforms and a database of labeled information of the transmission line. Based on the fusion improvement synthetic minority over-sampling technique (SMOTE) algorithm for sample equalization, the time domain function, fast Fourier transform (FFT), wavelet packet analysis, waveform function are used to extract the fault characteristics of different transmission line fault types in the time domain, frequency domain, time-frequency domain, and the waveform pulse width. The feature selection is completed based on deep least absolute shrinkage and selection operator (Deep Lasso) regression, and then, multi-source information fusion is finished. Finally, KOA-CNN-BiLSTM-Selfattention algorithm is adopted to complete the fault type identification of transmission lines. The correctness and effectiveness of the proposed method are verified by an example based on the historical fault tripping data of transmission lines.