基于加速卷积神经网络的变压器差动保护算法  

Transformer differential protection algorithm based on accelerated convolutional neural network

作　　者：茹瑞鹏 马建华[2] RU Ruipeng;MA Jianhua

机构地区：[1]山西潞安矿业集团慈林山煤业有限公司李村煤矿,山西长治046204 [2]辽宁工程技术大学,辽宁葫芦岛125000

出　　处：《现代机械》2024年第4期83-88,共6页Modern Machinery

基　　金：国家自然科学基金项目资助(项目编号51974151)。

摘　　要：针对变压器差动保护存在因励磁涌流而误跳闸的问题,提出一种基于加速卷积神经网络的算法。采用神经网络来区分内部故障电流和浪涌电流,应用压缩全连接层和卷积层并集成修正线性单元激活函数和批量归一化技术。用PSCAD/EMTDC软件建立220 kV变压器差动保护模型并应用算法,验证了算法更快、更可靠。Aiming at the problem of false tripping due to excitation surge current in transformer differential protection,an algorithm based on accelerated convolutional neural network is proposed.A neural network is used to distinguish between internal fault current and surge current,the compression of fully connected and convolutional layers is applied,and the modified linear cell activation function and batch normalization technique are integrated.The 220 kV transformer differential protection model is established with PSCAD/EMTDC software and the algorithm is applied,which verifies that the algorithm is faster and more reliable.

关 键 词：差动保护 电力变压器 浪涌电流 加速卷积神经网络 

分 类 号：TM4[电气工程—电器]