基于并联卷积神经网络的多端直流输电线路故障诊断

Fault Diagnosis of Multi-terminal HVDC Transmission Line Based on Parallel Convolutional Neural Network

针对多端直流输电(MTDC)线路故障时存在故障电流上升速度快、峰值大、不易定位等特点,提出一种兼顾快速性与准确性的MTDC线路故障诊断方法。首先,分析MTDC线路故障信号波形的幅值特征和频率特征,研究基于信号波形幅值变化的故障幅值特征提取方法和基于小波包分析的故障频率特征提取方法,进而形成基于幅值-频率特征的MTDC线路故障诊断方法。其次,构建具有故障分类支路和故障定位支路的双支路结构卷积神经网络--并联卷积神经网络(PCNN),提出基于迁移学习的P-CNN训练方法。最后,仿真验证基于P-CNN的MTDC线路故障诊断方法满足故障诊断的快速性要求,且其并联结构相比于其他人工智能故障诊断方法更具有准确性和可拓展性。

In view of the fault characteristics of multi-terminal high voltage direct current(MTDC)transmission lines,such as rapid rising speed,high fault peak current and difficulty in fault location,a fault diagnosis method with both rapidity and accuracy is proposed for MTDC system.Firstly,the amplitude and frequency characteristics of fault signal waveforms of MTDC transmission line faults are analyzed.The fault amplitude and frequency characteristic extraction methods are studied,based on amplitude variation of signal waveforms and wavelet packet analysis,respectively.Then,the fault diagnosis method of MTDC transmission system based on amplitude-frequency characteristics is obtained.Secondly,the parallel convolutional neural network(P-CNN)with fault classification and fault location dual branch structure is constructed,and the training method of P-CNN based on transfer learning is proposed.Finally,the simulation verifies that the fault diagnosis method of MTDC system based on P-CNN can meet the speed requirement,and the parallel structure is more accurate and expandable than other artificial intelligence fault diagnosis methods.