基于CNN_LSTM模型的复杂支路故障电弧检测

Fault arc detection of complex branch based on CNN_LSTM model

在低压交流配电系统中,当多支路并联的复杂系统的某1支路中出现串联电弧故障时,识别难度大幅提升。为了预防此类情况引发的电气火灾,提出1种卷积神经网络(CNN)与长短时记忆网络(LSTM)结合的串联故障电弧检测方法。首先,搭建实验平台用以采集不同负载在不同支路下发生故障时和正常工作时的干路电流数据;然后,构建CNN_LSTM模型并做出相应改进,将电流数据直接输入到模型中,由模型自主提取波形特征并进行分类。研究结果表明:该方法可以快速、准确地识别出电弧故障,准确率达99.04%以上,且能够较为准确地检测出是哪类负载所在的支路发生电弧故障,准确率达97.90%,可为复杂支路下的电弧故障识别研究提供参考。

In the low voltage AC distribution system,the difficulty of identification is greatly increased when the series arc fault occurs in one branch of the complex system with multiple branches in parallel.To prevent the electrical fires caused by such conditions,a detection method of series fault arc combining with the convolutional neural network(CNN)and long short-time memory network(LSTM)was proposed.Firstly,an experimental platform was built to collect the data of the trunk circuit current of different loads under different branches at fault and normal operation.Then the CNN_LSTM model was built and improved accordingly.The current data was directly input into the model,and the waveform features were extracted and classified by the model independently.The results showed that this method could quickly and accurately identify the arc faults,and the accuracy reached more than 99.04%.Moreover,it could more accurately detect the branch where the arc fault occurred with which kind of load,with the accuracy of 97.90%.It provides reference for the research of arc fault identification under complex branches.