随着新能源技术的发展，越来越多的逆变型分布式电源（Inverter Interfaced Distributed Generation，IIDG）并入配电网，虽然有效的缓解了我国的能源危机，但也使得配电网的故障区段定位问题更加复杂。由于IIDG输出的故障暂态电流受控制方式和锁相环等因素的影响呈现非线性的特征，现有的基于暂态信号的区段定位方法存在故障判据难以整定的问题。为此，本文利用残差网络（Residual Network，ResNet）能够在训练过程中自主精准整定判据的特点，提出了一种基于改进ResNet网络的含IIDG配电网故障区段定位方法。本文主要工作如下：

（1）在分析IIDG故障输出暂态特性基础上，研究了含IIDG配电网发生故障的故障暂态特征。研究表明，故障区段的两侧故障相电流暂态波形存在差异，非故障区段两侧暂态电流波形基本相同。仿真结果表明，上述特征不受故障位置和故障类型的影响；这为故障区段定位方法提供了理论支持。

（2）针对将改进的ResNet网络应用至含IIDG配电网故障区段定位中需要解决的问题，即一维电流数据不能直接作为改进ResNet网络输入和配电网故障样本数量不足导致改进的ResNet网络识别故障区段的精度不高，本文提出了基于故障灰度图的样本构造方法和基于改进深度卷积对抗生成网络（Deep Convolutional Generative Adversarial Network，DCGAN）的样本扩充方法。前者将各个检测点的三相时序电流数据按行拼接为二维数据矩阵，并通过灰度编码转换为二维故障灰度图，以此构建了改进ResNet网络的输入样本。后者首先对传统DCGAN进行改进，引入了条件信息和最小二乘损失函数来优化生成图像的质量。其次将故障区段的标签作为条件信息与已有样本输入改进的DCGAN，网络生成对应故障区段的故障灰度图，实现配电网故障样本的扩充。

（3）针对现有基于暂态信号的方法存在故障判据难以整定的问题，本文提出了一种基于改进ResNet网络的区段定位方法。首先对传统ResNet网络进行改进，采用多尺度卷积核替换传统残差模块的卷积核以减小计算量和参数量；同时添加了自注意力机制优化网络提取样本特征的权重，以提高网络的准确性。其次，通过样本集训练改进的ResNet网络，网络在训练过程中自主提取样本集的特征，实现判据的自主整定；故障发生时，将各区段两端采集到的时序电流数据转化为故障灰度图，并输入训练完毕的网络，网络根据自主整定的判据识别对应的故障区段，实现故障区段定位。仿真结果表明，本文方法可以准确地定位故障区段，并对多种干扰情况具有良好的适应性。

With the development of new energy technology, more and more Inverter Interfaced Distributed Generation (IIDG) are integrated into the distribution network, which effectively alleviates the energy crisis in China, but also makes the fault zone location problem of the distribution network more complicated. Since the fault transient current output from IIDG is nonlinear due to the control mode and phase-locked loop, the existing transient signal-based zone location method has the problem that the fault criterion is difficult to be rectified. In this paper, we propose an improved ResNet network-based fault location method for distribution networks with IIDG by using the feature of Residual Network (ResNet) that can autonomously and accurately adjust the criterion during the training process. The main work of this paper is as follows:

(1) Based on the analysis of the transient characteristics of the IIDG fault output, the fault transient characteristics of faults occurring in the distribution network containing IIDG are studied. The study shows that there are differences in the transient waveforms of the fault phase currents on both sides of the fault section, and the transient current waveforms on both sides of the non-fault section are basically the same. The simulation results show that the above characteristics are not affected by the fault location and fault type; this provides theoretical support for the fault segment location method.

(2) To address the problems that need to be solved in applying the improved ResNet network to the fault segment location of distribution networks containing IIDG, i.e., the one-dimensional current data cannot be directly used as the input of the improved ResNet network and the insufficient number of fault samples in the distribution network leads to the low accuracy of the improved ResNet network in identifying the fault segments, this paper proposes a sample construction method based on the fault gray map and an improved In this paper, we propose a sample construction method based on the fault gray map and a sample expansion method based on the improved Deep Convolutional Generative Adversarial Network (DCGAN). The former constructs the input samples of the improved ResNet network by stitching the three-phase timing current data of each detection point into a two-dimensional data matrix by row and converting it into a two-dimensional fault gray map by grayscale coding. The latter firstly improves the traditional DCGAN by introducing conditional information and least-squares loss function to optimize the quality of the generated images. Secondly, the labels of the fault segments are used as the conditional information and the existing samples are input to the improved DCGAN, and the network generates the fault grayscale maps of the corresponding fault segments to realize the expansion of the distribution network fault samples.

(3) To address the problem that the existing transient signal-based methods have fault criteria that are difficult to rectify, this paper proposes a zone location method based on the improved ResNet network. Firstly, the traditional ResNet network is improved by replacing the convolutional kernel of the traditional residual module with a multi-scale convolutional kernel to reduce the computation and the number of parameters; at the same time, a self-attention mechanism is added to optimize the weights of sample features extracted by the network to improve the accuracy of the network. Secondly, the improved ResNet network is trained with the sample set, and the network extracts the features of the sample set during the training process to realize the independent adjustment of the criterion; when a fault occurs, the time series current data collected at both ends of each segment are transformed into the grayscale map of the fault and input to the trained network, and the network identifies the corresponding fault segment according to the independent adjusted criterion to realize the fault segment location. The simulation results show that this method can accurately locate fault segments and has good adaptability to various interference situations.

[1]詹惠瑜, 刘科研, 盛万兴, 等. 有源配电网故障诊断与定位方法综述及展望[J]. 高电压技术, 2023, 49(02): 660-671.

[2]Wang XL, He CC, Li XM, et al. The vulnerability analysis of distribution network with distributed Generation[J]. Electrical Measurement & Instrumentation, 2019, 56(06): 38-43.

[3]李卫彬, 童欣, 黄超, 等. 基于分层定位模型的含DG配电网故障定位方法研究[J]. 电力系统保护与控制, 2022, 50(24): 160-166.

[4]贾科, 郑黎明, 毕天姝, 等. 基于余弦相似度的风电场站送出线路纵联保护[J]. 中国电机工程学报, 2019, 39(21): 6263-6275.

[5]贾科, 杨哲, 魏超, 等. 基于斯皮尔曼等级相关系数的新能源送出线路纵联保护[J]. 电力系统自动化, 2020, 44(15): 103-111.

[6]Mohamed M.A. Mahfouz. A protection scheme for multi-distributed smart microgrid based on auto-cosine similarity of feeders current patterns[J].Electric Power Systems Research,2020,186(4):167-169.

[7]魏东辉, 于舜尧, 房俊龙. 基于综合序电流的含光伏电源配电网纵联保护方案[J]. 太阳能学报, 2021, 42(07): 185-192.

[8]周成瀚, 邹贵彬, 杜肖功, 等. 基于正序电流故障分量的有源配电网纵联保护[J]. 中国电机工程学报, 2020, 40(07): 2102-2112+2390.

[9]和敬涵, 李文立, 张大海, 等. 基于节点分支电流幅值的含DG配网差动保护原理[J]. 电网技术, 2018, 42(11): 3601-3609.

[10]马静, 刘静. 基于故障稳态分量的含DG配电网自适应方向电流保护方案[J]. 电力自动化设备, 2018, 38(01): 1-9.

[11]韩博文, 王钢, 李海锋, 等. 含逆变型分布式电源配电网的新型纵联保护方案[J]. 高电压技术, 2017, 43(10): 3453-3462.

[12] Chen XL, Yuan S, Li YL, et al. Fault location method for petal-shaped distribution network with inverter-interfaced distributed generators[J]. Global Energy Interconnection, 2021, 4(6): 2096-5117.

[13]高岩, 李永丽, 陈晓龙, 等. 基于电流幅值比的有源配电网自适应差动保护原理[J]. 电力系统及其自动化学报, 2021, 33(02): 1-7.

[14]王子璇, 马啸, 杨勇, 等. 计及不可测分支负荷电源助增效应的有源配网幅值差动保护新判据[J]. 中国电机工程学报, 2020, 40(S1): 56-68.

[15]董思柯. 含分布式电源的配电网线路保护研究[D]. 西安：西安科技大学, 2020.

[16]Zhao JW, Zhao RJ, Shi W, et al. Fault Location Based on Comparison of Current Information for Distribution Network with DGs[C]. The 2019 IEEE 3rd International Electrical and Energy Conference (CIEEC2019), 2019: 7-9.

[17]李佼洁, 刘毅力. 基于电流相角变化量的含DG配网保护方案研究[J]. 国外电子测量技术, 2021, 40(09): 6-12.

[18]李瀚霖, 靳维, 梁睿, 等. 基于故障信息自同步的有源配电网纵联保护[J]. 电力工程技术, 2021, 40(06): 121-126+133.

[19]牛耕, 周龙, 裴玮, 等. 低压有源配电网在线故障区间定位与识别方法[J]. 中国电机工程学报, 2017, 37(09): 2525-2539.

[20]杨国华, 冯骥, 柳萱, 等. 基于改进秃鹰搜索算法的含分布式电源配电网分区故障定位[J]. 电力系统保护与控制, 2022, 50(18): 1-9.

[21]赵冰杰, 贾宏杰, 李沐阳, 等. 基于分组差分粒子群算法的含分布式电源配电网故障定位[J]. 可再生能源, 2022, 40(10): 1380-1386.

[22]任志玲, 刘卫东, 杨柳, 等. 基于改进鸽群算法的含分布式电源配电网故障定位[J]. 电源学报, 2022, 20(04): 171-178.

[23]胡福年, 张认, 卞小亮. 基于图论的主动配电网故障定位方法[J]. 控制工程, 2021, 28(05): 833-838.

[24]和敬涵, 王紫琪, 张大海. 基于图论及模糊评价的广域后备保护分区方法[J]. 电力自动化设备, 2017, 37(02): 75-82.

[25]廖碧莲, 高中林, 唐江琦, 等. 基于图论的主动配电网集中式方向保护研究[J]. 智慧电力, 2019, 47(08): 55-62.

[26]晁晨栩, 郑晓冬, 高飘, 等. 含高比例光伏配电网的高频阻抗差动保护[J]. 中国电机工程学报, 2021, 41(20): 6968-6979.

[27]Li X, Lu Y. High Frequency Impedance Based Fast Directional Pilot Protection Scheme for Distribution Network With Wind DGs[C]. 2020 12th IEEE PES Asia-Pacific Power and Energy Engineering Conference (APPEEC), 2020: 1-6.

[28]Liu S, Zhang LL, Fu C, et al. A New Two-Port Network Model-Based Pilot Protection for AC Transmission Lines [J]. IEEE Transactions on Power Delivery, 2020, 35(2): 473-482.

[29]Wang J, Li Y, Cost P, et al. A novel pilot protection for transmission lines based on full travelling wave signal polarities[C]. 15th International Conference on Developments in Power System Protection (DPSP 2020), 2020: 1-6.

[30]许名扬, 樊艳芳, 刘群杰, 等. 基于电流初始行波极性的主动配电网纵联 保护方案[J]. 可再生能源, 2020, 38(04): 509-516.

[31]宋国兵, 王晨清, 唐吉斯, 等. 适用于风电接入系统的时域模型识别纵联 保护新原理[J]. 电网技术, 2016, 40(11): 3580-3585.

[32]李育强, 宋国兵, 王维庆, 等. 基于模型识别光伏接入配电网线路纵联保 护研究[J]. 电力系统保护与控制, 2016, 44(06): 98-103.

[33]王春又, 孙士云, 毛肖, 等. 适应于双馈风电场送出线的时域距离纵联方 向保护[J]. 电力系统保护与控制, 2021, 49(13): 82-94.

[34]Zhang G, Tong X, Hong Q, et al. Waveform Similarity-Based Robust Pilot Protection for Transmission Lines[J]. IEEE Transactions on Power Delivery, 2021: 1-9 .

[35]Yang Z, Liao W, Wang H, et al. Improved Euclidean Distance Based Pilot Protection for Lines with Renewable Energy Sources[J]. IEEE Transactions on Industrial Informatics, 2022: 1-10 .

[36]刘志坚, 何蔚, 刘航, 等. 基于格拉姆角场变换和深度压缩模型的变压器故障识别方法[J/OL]. 电网技术: 1-13[2023-04-09].

[37]郑炜, 林瑞全, 王俊, 等. 基于GAF与卷积神经网络的电能质量扰动分类[J]. 电力系统保护与控制, 2021, 49(11): 97-104.

[38]毛向向, 王红军, 韩凤霞, 等. 基于深度卷积神经网络的机电系统故障分类识别方法[J]. 电子测量与仪器学报, 2021, 35(02): 87-93.

[39]金岩磊, 何茂慧, 郭涛, 等. 改进VMD融合深度学习在滚动轴承故障诊断中的应用[J]. 热能动力工程, 2023, 38(02): 144-152.

[40]孙晔, 张泽明, 刘晓悦. 基于ResNet-ELM和迁移学习的风机齿轮箱故障诊断方法研究[J/OL]. 机电工程: 1-12[2023-04-09].

[41]袁科研, 兰媛, 黄家海, 等. 基于GADF和ResNet的轴向柱塞泵复合故障诊断研究[J/OL]. 机电工程: 1-8[2023-04-09].

[42]毕天姝, 李彦宾, 贾科, 等. 基于暂态电流波形相关性的新能源场站送出线路纵联保护[J]. 中国电机工程学报, 2018, 38(07): 2012-2019+2216.

[43]Chang Z, Song G, Wang T. Analysis on current characteristics of PMSG under grid three-phase fault[J]. The Journal of Engineering, 2018, 15(10): 785-790.

[44]孔祥平, 张哲, 尹项根, 等. 含逆变型分布式电源的电网故障电流特性与故障分析方法研究[J]. 中国电机工程学报, 2013, 33(34): 65-74, S9.

[45]国家电网公司. 分布式电源接入电网技术规定: Q/GDW 1480-2015[S]. 2016.

[46]赵建文, 李璞, 冯泽镡. 基于AGM曲线能量的配电网故障区段定位方法[J]. 电力系统及其自动化学报, 2021, 33(04): 108-114+126.

[47]Verband der Netzbetreiber VDN e. v. beim VDEW. Network and system rules of the German transmission system operators[EB/OL]. Berlin, German: VDN, 2007 [2012-9-23].

[48]刘鹏辉. 基于信号特征辨识的配电网馈线保护方法研究[D]. 长沙:湖南大学, 2019.

[49]陈雪洁. 中压电网单相接地故障特征及选线研究[D]. 西安: 西安科技大学, 2012.

[50]罗溢, 李开成, 肖贤贵, 等. 基于马尔可夫转移场和深度残差网络的电能质量复合扰动多标签分类[J/OL]. 中国电机工程学报: 1-11[2023-04-09].

[51]张淑清, 杜灵韵, 王册浩, 等. 基于格拉姆角场与改进CNN-ResNet的风电功率预测方法[J/OL]. 电网技术: 1-10[2023-04-09].

[52]陈勤, 伍传敏. 基于主成分分析的彩色图转灰度图方法的研究[J]. 佳木斯大学学报(自然科学版), 2018, 36(03): 380-383.

[53]岳震. 基于深度学习的有源配电网故障定位与预测研究[D]. 北京：华北电力大学(北京), 2021.

[54]刘云鹏, 李瑾, 王宇, 蔡文立, 等. 图像增强下基于生成对抗网络和卷积神经网络的CT与MRI融合方法[J/OL]. 生物医学工程学杂志: 1-9[2023-04-09].

[55]张逸骋, 井花花, 晏涛. 基于条件生成对抗网络的光场图像透视视图生成算法[J/OL]. 计算机应用研究: 1-8[2023-04-09].

[56]尹玉婷, 肖秦琨. 基于深度卷积生成对抗网络的图像生成[J]. 计算机技术与发展, 2021, 31(04): 86-92.

[57]杨思渊, 蒋锐鹏, 海仁古丽·阿不力提甫, 等. 基于小波深度卷积生成对抗网络的人脸图像生成[J]. 信息通信, 2020(11): 69-72.

[58]王欣, 付威. 基于推土机距离的证据冲突强度量方法[J]. 通信学报, 2022, 43(05): 204-213.

[59]李亚舟, 曹江涛, 姬晓飞. 融合改进Padim建模和ResNet网络的喷涂质量检测算法[J]. 电子测量与仪器学报, 2022, 36(11): 91-97.

[60]朱传军, 刘荣光, 成佳闻, 等. 基于SimAM模块与ResNet34网络的混合缺陷检测模型[J]. 现代制造工程, 2023(02): 1-9.

[61]朱林. 基于实时以太网的数字化变电站体系结构和新型应用[D]. 保定: 华中科技大学, 2010.