基于深度神经网络的UHVDC输电系统故障诊断

Fault Diagnosis of UHVDC Transmission Lines Based on Deep Neural Network

针对传统特高压直流(UHVDC)故障诊断方法存在阈值整定复杂、灵敏度低以及耐受过渡电阻能力较弱的问题,提出了一种将多尺度卷积神经网络(multi-scale convolutional neural network,MCNN)、双向长短时记忆网络(bidirectional long short-term memory,BiLSTM)和注意力(Attention)机制相结合的UHVDC输电系统故障诊断方法。通过MCNN挖掘标准化后的故障数据不同尺度的空间特征;利用双层BiLSTM获取数据中的时序依赖特征;引入Attention机制为数据的不同特征向量合理分配注意力。结果表明:所提方法在4种评价指标上都优于其他对比算法,能够准确识别UHVDC输电系统的各种区内、外故障和测量故障,并且在面对高阻故障时仍然具有较高的分类精度。

In view of the complex threshold setting,low sensitivity and weak ability to withstand transition resistance in traditional UHVDC fault diagnosis methods,a fault diagnosis method for the UHVDC transmission system combining multiscale convolutional neural network(MCNN),bidirectional long short-term memory(BiLSTM)and attention mechanism is proposed in this paper.Firstly,the spatial features of different scales of standardized fault data are mined by MCNN.Secondly,the time-dependent features in the data are obtained by using double-layer BiLSTM.Finally,the attention mechanism is introduced to reasonably allocate attention to different feature vectors of the data.The findings demonstrate that the proposed method outperforms other algorithms in four evaluation indexes,and can accurately discern all kinds of internal and external faults and measurement faults of the UHVDC transmission system,and still has high classification accuracy when facing high impedance faults.