基于卷积双向长短期记忆网络的微网继电保护故障诊断技术  

FAULT DIAGNOSIS TECHNOLOGY OF RELAY PROTECTION IN MICROGRID BASED ON CONVOLUTIONAL BIDI-RECTIONAL LONG SHORT-TERM MEMORY NETWORK

作　　者：杨志淳 闵怀东 杨帆 雷杨 胡伟 陈鹤冲 Yang Zhichun;Min Huaidong;Yang Fan;Lei Yang;Hu Wei;Chen Hechong(State Grid Hubei Electric Power Research Institute,Wuhan 430077,China)

机构地区：[1]国网湖北省电力有限公司电力科学研究院,武汉430077

出　　处：《太阳能学报》2025年第1期420-428,共9页Acta Energiae Solaris Sinica

基　　金：国网湖北省电力有限公司电力科学研究院科技项目(SGHBDK00PWJS2200077)。

摘　　要：分布式电源种类和容量不断提升的微网运行方式复杂、故障特征微弱,现有的继电保护装置故障诊断方法无法满足保护需求。提出一种基于卷积双向长短期记忆网络的微网继电保护故障诊断技术。首先,分析多能源互补微网系统架构,对采集的三相电流数据进行预处理,提高后续模型对数据的学习效率;然后,融合卷积神经网络和双向长短期记忆网络提出卷积双向长短期记忆网络的微网继电保护故障诊断方法,提取三相电流数据长序列和局部序列特征实现故障分类、故障定位,融合注意力机制,重点关注对故障诊断有影响的特征,提高故障诊断准确率;最后经过RTDS实时仿真系统进行验证,实验结果表明,所提方法故障诊断精度高、计算时间短,同卷积神经网络、长短期记忆网络、人工神经网络相比,故障分类准确率分别提升8.53%、9.62%、11.45%,故障定位准确率分别提升7.47%、10.61%、10.85%,验证所提方法的有效性与先进性。The operation mode of microgrids with continuously increasing types and capacities of distributed power sources is complex and the fault characteristics are weak.The existing fault diagnosis methods for relay protection devices cannot meet the protection requirements.A fault diagnosis technique for relay protection in microgrid based on convolutional bidirectional Long short-term memory network is proposed.First,analyze the architecture of the multi energy complementary microgrid system,preprocess the collected three-phase current data,and improve the learning efficiency of the subsequent model on the data;Then,combining Convolutional neural network and bi-directional Long short-term memory network,a fault diagnosis method for micro network relay protection based on convolutional bidirectional Long short-term memory network is proposed.Long sequence and local sequence features of three-phase current data are extracted to achieve fault classification and fault location.The fusion attention mechanism focuses on features that have an impact on fault diagnosis,and improves the accuracy of fault diagnosis;Finally,the RTDS real-time simulation system is used to verify the experimental results.The experimental results show that the proposed method has high fault diagnosis accuracy and short calculation time.Compared with Convolutional neural network,long short-term memory network and artificial neural network,the accuracy of fault classification is increased by 8.53%,9.62%and 11.45%respectively,and the accuracy of fault location is increased by 7.47%,10.61%and 10.85%,which verifies the effectiveness and progressiveness of the proposed method.

关 键 词：微网 继电保护 故障诊断 卷积双向长短期记忆网络 三相电流 注意力机制 

分 类 号：TM732[电气工程—电力系统及自动化]