电缆短路是煤矿电网中危害严重、发生频次较高的电气故障。不同于公用配电网，受矿井恶劣电磁环境以及企业电能质量不高等影响，煤矿电缆短路类型多，其故障特性具有较强非线性、不稳定性和分散性特点，导致短路故障识别及其类型判定困难，易造成保护出现越级跳闸或拒动现象，而且越级跳闸后故障原因排查困难，已成为煤矿供电安全的突出问题，而提高短路故障识别准确率及其类型判定精度是解决上述问题的关键。

      本文在分析煤矿电缆短路故障特性以及电气扰动的基础上，针对现有方法在识别煤矿电缆短路故障时存在无法挖掘故障深层特征和泛化性能弱等不足，将基于深度学习的智能识别方法用于煤矿电缆短路故障研究中，设计了适用于煤矿电缆短路故障识别及其类型智能判定的方案。

       由于现有方法需凭借先验知识选取输入特征量，导致短路故障深层特征难以提取，本文利用深度学习可深入挖掘并自动提取故障深层特征的思想，提出一种基于堆栈自编码器的短路故障特征提取方法。该方法在对故障数据分解与重构的基础上，自动从原始故障数据中获取深层特征，解决了短路故障的强非线性、不稳定性而导致现有方法故障特征提取困难的问题，为准确识别短路故障并判定其类型奠定基础。

针对深度学习方法用于煤矿电缆短路故障识别时存在的过拟合问题，本文采用具有集成思想的Dropout技术改进堆栈自编码器模型，并采用Adam算法优化模型参数，避免了传统堆栈自编码器因训练集与测试集存在差异，导致训练集上短路故障分类效果良好而测试集分类准确率降低的问题；采用Min-Max归一化与拼接重构相结合的数据处理方法来提升故障数据与对应类型之间的关联性，有效提升了模型在煤矿复杂工况环境下的稳定性与鲁棒性。

       深度学习方法在解决分类问题时通常采用Softmax分类器，为了克服Softmax分类器对特征差异性小的故障类型判定能力不强的缺陷，采用鲁棒性强的极限学习机替代Softmax分类器，充分保留与扩大了不同特征之间的差异性，以提高本文模型对与短路故障特征相似的强电气扰动的辨识能力。

       分别以煤矿电网实际运行参数为依据的仿真模型和现场实测数据进行实验分析。利用Loss曲线、T-分布随机近邻嵌入算法与混淆矩阵对本文方法进行可视化分析，在随机故障发生条件与不同故障类型下将本文方法与现有方法进行对比，并采用准确率与精度指标定量分析短路故障判识结果，验证了本文方法能够在煤矿复杂工况环境下精确判定短路故障类型，故障识别准确率相对于传统SAE模型提高了11.4%。

Cable short circuit is a serious and frequent electrical fault in coal mine power grid. Different from the public distribution network, due to the harsh electromagnetic environment of the mine and the low power quality of the enterprise, there are many types of short-circuit faults in coal mine cables, and their fault characteristics have strong non-linearity, instability and dispersion characteristics, which makes it difficult to identify and determine the types of short-circuit faults. It is easy to cause the phenomenon of override trip or rejection of protection. It is difficult to investigate the causes of faults after override trip, which has become a prominent problem of coal mine power supply safety. Improving the accuracy of short-circuit fault identification and its type determination accuracy is the key to solve the above problems.

Based on the analysis of the characteristics of short-circuit fault of coal mine cable and electrical disturbance, this thesis aims at the shortcomings of the existing methods in identifying the short-circuit fault of coal mine cable, such as the inability to excavate the deep characteristics of the fault and the weak generalization performance. The intelligent identification method based on deep learning is used in the study of short-circuit fault of coal mine cable, and a scheme suitable for short-circuit fault identification of coal mine cable and intelligent determination of its type is designed.

Since the existing methods need to select the input feature quantity by virtue of prior knowledge, it is difficult to extract the deep features of short-circuit faults. This thesis uses the idea that deep learning can deeply mine and automatically extract the deep features of faults, and proposes a short-circuit fault feature extraction method based on stacked autoencoder. Based on the decomposition and reconstruction of fault data, this method automatically obtains deep features from the original fault data, solves the problem that the strong nonlinearity and instability of short-circuit fault lead to the difficulty of fault feature extraction in existing methods, and lays a foundation for accurately identifying short-circuit fault and determining its type.

Aiming at the over-fitting problem of deep learning method for short-circuit fault identification of coal mine cables, this thesis uses Dropout technology with integrated idea to improve the stack auto-encoder model, and uses Adam algorithm to optimize the model parameters, which avoids the problem that the traditional stack auto-encoder has a good classification effect of short-circuit fault on the training set and a low classification accuracy of the test set due to the difference between the training set and the test set. The data processing method combining Min-Max normalization and splicing reconstruction is used to improve the correlation between fault data and corresponding types, which effectively improves the stability and robustness of the model under complex working conditions in coal mines.

The deep learning method usually uses the Softmax classifier to solve the classification problem. In order to overcome the defect that the Softmax classifier has weak ability to determine the fault type with small feature difference, the robust extreme learning machine is used to replace the Softmax classifier, which fully retains and expands the difference between different features, so as to improve the identification ability of the model in this thesis to strong electrical disturbances similar to short-circuit fault features.

The model based on the actual operating parameters of the coal mine power grid and the field measured data are analyzed experimentally. The Loss curve, T-distributed random neighbor embedding algorithm and confusion matrix are used to visually analyze the proposed method. The proposed method is compared with the existing methods under random fault occurrence conditions and different fault types, and the accuracy and accuracy indicators are used to quantitatively analyze the short-circuit fault identification results. It is verified that the proposed method can accurately determine the type of short-circuit fault in the complex working environment of coal mine, and the accuracy of fault identification is improved by 11.4 % compared with the traditional SAE model.

[1] 韩颖, 李向威, 闫志佳. 我国高温矿井热害防治技术研究进展及展望[J]. 煤炭技术, 2023, 42(10): 186-189.

[2] 王海军, 曹云, 王洪磊. 煤矿智能化关键技术研究与实践[J]. 煤田地质与勘探, 2023, 51(01): 44-54.

[3] 李小敏. 基于暂态信号分析的矿井电网单相漏电故障选线[J]. 国外电子测量技术, 2023, 42(02): 164-170.

[4] Zhang Baohui, Hao Zhiguo, Bo Zhiqian. New development in relay protection for smart grid[J]. Protection and Control of Modern Power Systems, 2016, 1(1): 121-127.

[5] 罗娟, 贠保记, 雷富坤, 等. 防止煤矿供电越级跳闸新方法的研究[J]. 西安科技大学学报, 2014, 34(01): 62-66.

[6] 王永军, 刘鑫. 基于信号注入和分散检测的井下供电系统防越级研究[J]. 中国安全科学学报, 2023, 33(S1): 76-80.

[7] 张冠英, 张晓亮, 刘华, 等. 低压系统串联故障电弧在线检测方法[J]. 电工技术学报, 2016, 31(08): 109-115.

[8] 赖逸洋, 王增平, 王彤.电流差动保护在柔直接入的交流电网中适应性分析及改进措施研究[J]. 电力系统保护与控制, 2023, 51(03): 145-154.

[9] 张太升, 韩伟, 杨霖, 等. 基于波形相关分析的双馈风电场送出线时域距离保护[J].电力系统保护与控制, 2021, 49(14): 82-88.

[10] 朱晓娟, 林圣, 张姝, 等. 基于小波能量矩的高阻接地故障检测方法[J]. 电力自动化设备, 2016, 36(12): 161-168.

[11] 余南华, 李传健, 杨军, 等. 基于小波包时间熵的配电网运行状态特征提取方法[J]. 电力自动化设备, 2014, 34(09): 64-71+79.

[12] 邹贵彬, 高厚磊, 王明军, 等. 一种雷电波侵入变电站的扰动识别方法[J]. 电网技术, 2012, 36(04): 179-183.

[13] 姚海燕, 张静, 留毅, 等. 基于多尺度小波判据和时频特征关联的电缆早期故障检测和识别方法[J]. 电力系统保护与控制, 2015, 43(09): 115-123.

[14] 杨毅, 范栋琛, 殷浩然, 等. 基于深度-迁移学习的输电线路故障选相模型及其可迁移性研究[J]. 电力自动化设备, 2020, 40(10): 165-172.

[15] 唐冠宏, 黄潘永. 改进决策树对10kV配电网分支线断线故障诊断与定位的优化分析[J]. 电工技术, 2021, (10): 129-132.

[16] Asman S H, Aziz N F A, Amirulddin U. Decision tree method for fault causes classification based on RMS-DWT analysis in 275kV transmission lines network[J]. Apply Science, 2021, 11(09): 4031-4051.

[17] Dehbozorgi M, Rastegar M, Dabbaghjamanesh M. Decision tree-based classifiers for root-cause detection of equipment-related distribution power system outages[J]. IET Generation, Transmission&Distribution, 2020, 14(24): 5809-5815.

[18] 李胜辉, 白雪, 董鹤楠, 等. 基于平稳小波变换与随机森林的电缆早期故障识别方法[J]. 电工电能新技术, 2020, 39(03): 40-48.

[19] 杨杰, 吴浩, 董星星, 等. 基于电流故障分量特征和随机森林的输电线路故障类型识别[J]. 电力系统保护与控制, 2021, 49(13): 53-63.

[20] 白浩, 潘姝慧, 邵向潮, 等. 基于小波去噪与随机森林的配电网高阻接地故障半监督识别方法[J]. 电力系统保护与控制, 2022, 50(20): 79-87.

[21] Torres J O A, Santos R C D. An ANN-based algorithm for fault detection and location in MTDC systems[C]//Proceedings of 2018 Simposio Brasileiro de Sistemas Eletricos. Niteroi, Brazil: IEEE, 2018: 1-6.

[22] Ngaopitakkul A, Pothisarn C, Bunjongjit S, et al. DWT and RBF neural networks algorithm for identifying the fault types in underground cable[C]//TENCON 2011-2011 IEEE Region 10 Conference. IEEE, 2011: 1379-1382.

[23] 韩泰然. 基于神经网络专家系统煤矿高压电缆故障诊断策略[J]. 煤炭技术, 2019, 38(11): 177-180.

[24] 谢国民, 黄睿灵, 丁会巧. 基于VMD样本熵和KELM的输电线路故障诊断[J]. 电子测量与仪器学报, 2019, 33(05): 73-79.

[25] 王毅, 李曙, 李松浓, 等. 瞬时特征下极限学习机在接地故障诊断中的应用[J]. 电子测量与仪器学报, 2022, 36(01): 212-219.

[26] 黄南天, 李富青, 王文婷, 等. 输电线路故障层次化变步长Tsallis小波奇异熵诊断方法[J]. 电力系统保护与控制, 2017, 45(18): 38-44.

[27] Stollenga M F, Masci J, Gomez F, et al. Deep networks with internal selective attention through feedback connections[C]//Advances in Neural Information Processing Systems. 2014: 3545-3553.

[28] 黄景林, 彭显刚, 简胜超, 等. 基于深度学习与不平衡样本集的输电线路故障分类[J].智慧电力, 2021, 49(02): 114-119.

[29] 王晓东, 张昊, 郭海宇, 等. 光伏电站直流输电线路早期绝缘故障识别与定位[J]. 太阳能学报, 2023, 44(08): 246-252.

[30] 魏东, 龚庆武, 来文青, 等. 基于卷积神经网络的输电线路区内外故障判断及故障选相方法研究[J]. 中国电机工程学报, 2016, 36(S1): 21-28.

[31] 冯寅, 贾宏涛, 杨震强, 等. 基于CNN-LSTM的输电线路故障诊断方法研究[J]. 电网与清洁能源, 2023, 39(11): 59-65.

[32] 邓丰, 史鸿飞, 冯思旭, 等. CNN-LSTM全景故障特征挖掘的配电网单端定位方法[J]. 中国电机工程学报, 2023, 43(S1): 114-126.

[33] 谭媛, 张文海, 王杨. 基于多源信息融合的配网故障外部原因识别[J]. 电力建设, 2023, 44(03): 77-84.

[34] 周泽雷, 吴正平, 付翊航, 等. 多端直流电网的非边界依赖型线路故障辨识方案[J].电力系统及其自动化学报, 2023, 35(04): 100-108+123.

[35] 汪洋, 杨仕伟, 王宝华, 等. 基于深度置信网络的交直流配电网直流故障检测技术[J].电力工程技术, 2023, 42(01): 251-259.

[36] 汪颖, 卢宏, 杨晓梅, 等. 堆叠自动编码器与S变换相结合的电缆早期故障识别方法[J]. 电力自动化设备, 2018, 38(08): 117-124.

[37] 尚博阳, 罗国敏, 茹嘉昕, 等. 基于有限量测信息的多分支配电线路故障定位方法[J]. 高电压技术, 2023, 49(06): 2308-2319.

[38] 刁晓虹, 董昊, 侯磊, 等. 基于CEEMDAN-HT和SAE的直流故障电弧诊断方法研究[J]. 电网与清洁能源, 2023, 39(10): 1-8+18.

[39] 高云广. 矿井电网有源电力滤波器电流控制策略研究[D]. 太原: 太原理工大学, 2017.

[40] 彭迈. 矿用综合保护装置设计研究与应用[D]. 西安: 西安科技大学, 2019.

[41] 中华人民共和国应急管理部. 煤矿安全规程[M]. 北京: 应急管理出版社, 2022: 147-148.

[42] 张一, 张佰富, 韩肖清, 等. 直流侧电容中点接地的混合微电网故障传播机理[J]. 电工技术学报, 2023, 38(15): 4208-4220.

[43] Choi M S, Lee S J, Lee D S, et al. A newfault location algorithm using direct circuit analysis for distribution systems[J]. IEEE Transactions on Power Delivery, 2004, 19(01): 35-41.

[44] 阎治安. 电机学[M]. 陕西: 西安交通大学出版社. 2006: 171-172.

[45] 李振兴, 胡聪, 朱益, 等. 基于采样序列整形的配电网线路保护防励磁涌流误动策略[J]. 电力系统保护与控制, 2023, 51(21): 136-145.

[46] 朱乔木, 陈金富, 李弘毅, 等. 基于堆叠自动编码器的电力系统暂态稳定评估[J]. 中国电机工程学报, 2018, 38(10): 2937-2946+3144.

[47] 张汪洋, 樊艳芳, 侯俊杰, 等. 基于集成深度神经网络的配电网分布式状态估计方法[J]. 电力系统保护与控制, 2024, 52(03): 128-140.

[48] 杨玉萍, 吴浩, 田海鹏, 等. 高压直流输电线路单端智能故障定位方法[J]. 电力系统及其自动化学报, 2023, 35(09): 120-129.

[49] Wang Yixing, Liu Meiqin, Bao Zhejing, et al. Stacked sparse autoencoder with PCA and SVM for data-based line trip fault diagnosis in power systems[J]. Neural Computing and Applications, 2019, 31(10): 6719-6731.

[50] 叶林, 葛鸥翔, 郭永红, 等. 基于堆叠自动编码器的汽轮发电机多指标故障诊断模型[J]. 中国电机工程学报, 2022, 42(10): 3656-3670.

[51] Cho K, Courville A, Bengio Y. Describing multimedia content using attention-based encoder-decoder networks[J]. IEEE Transactions on Multimedia, 2015, 17(11): 1875-1886.

[52] 徐耀松, 李佳旺, 段彦强. 基于相似度机制AdaBoost-DBN的变压器故障层级诊断[J]. 高压电器, 2023, 59(06): 154-164.

[53] 张泽琳, 章智伟, 胡齐, 等. 基于深度学习的多产品煤料图像分类方法研究[J]. 煤炭科学技术, 2021, 49(09): 117-123.

[54] 陆云才, 廖才波, 李群, 等. 基于声纹特征和集成学习的变压器缺陷诊断方法[J]. 电力工程技术, 2023, 42(05): 46-55.

[55] 龚泽威一, 曹占国, 周帅, 等.基于过采样和随机森林的高压隔离开关发热故障诊断[J]. 电网与清洁能源, 2023, 39(11): 72-79+96.

[56] 杨建, 刘笑, 董密, 等.基于深度学习的恒功率负荷直流微电网稳定性分析[J]. 电力系统自动化, 2023, 47(15): 188-197.

[57] 崔峰, 何仕凤, 来兴平, 等. 基于相空间重构与深度学习的冲击地压矿井时间序列b值趋势[J]. 煤炭学报, 2023, 48(05): 2022-2034.

[58] 李东东, 刘宇航, 赵阳, 等. 基于改进生成对抗网络的风机行星齿轮箱故障诊断方法[J]. 中国电机工程学报, 2021, 41(21): 7496-7507.

[59] 张译丹, 李奇, 尹良震, 等. 引入自监督机制的燃料电池水管理系统健康状态检测方法[J]. 中国电机工程学报, 2023, 43(13): 5025-5037.

[60] Huang Guangbin, Zhu Qinyu, Siew C K. Extreme learning machine: Theory and applications[J]. Neurocomputing, 2005, 70(01): 489-501.

[61] Chen Zhicong, Wu Lijun, CHENG Shuying, et al. Intelligent fault diagnosis of photovoltaic arrays based on optimized kernel extreme learning machine and I-V characteristics[J]. Applied Energy, 2017, 204(15): 912-931.

[62] Shao Haidong, Jiang Hongkai, Li Xingqiu, et al. Intelligent fault diagnosis of rolling bearing using deep wavelet auto-encoder with extreme learning machine[J]. Knowledge Based Systems, 2018, 140(15): 1-14.

[63] 马莉, 潘少波, 代新冠, 等. 基于PSO-Adam-GRU的煤矿瓦斯浓度预测模型[J]. 西安科技大学学报, 2020, 40(02): 363-368.

[64] 曹洁, 张玉林, 王进花, 等. 基于VMD和SVPSO-BP的滚动轴承故障诊断[J]. 太阳能学报, 2022, 43(09): 294-301.

[65] 王征, 张科, 张赫林, 等. 深度可分离卷积神经网络miniXception对矿工情绪特征的识别[J]. 西安科技大学学报, 2022, 42(03): 562-571.

[66] 范敏, 杨青, 郭祥富, 等. 面向不平衡数据的配电网故障停电预测方法[J]. 电力系统保护与控制, 2023, 51(08): 96-106.