| 引用本文: | 林丰恺,王 建,赵 启,等.基于多模态残差网络融合波形与天气信息的输电线路故障原因辨识方法[J].电力系统保护与控制,2025,53(20):23-33.[点击复制] |
| LIN Fengkai,WANG Jian,ZHAO Qi,et al.Transmission line fault cause identification method based on multimodal residual network integrating waveform and weather information[J].Power System Protection and Control,2025,53(20):23-33[点击复制] |
|
| 摘要: |
| 针对现有基于暂态波形图像识别的输电线路故障原因辨识，因使用的输入特征相对单一而无法进一步细分故障原因的问题，基于多模态残差网络，提出了一种融合暂态波形与天气特征的输电线路故障原因辨识方法。首先，统计分析了不同原因的输电线路故障在暂态波形和天气条件方面的特征差异。其次，以暂态波形图像和故障发生时天气情况的独热编码作为改进多模态残差网络分类器的输入。然后，利用通道注意力机制对网络提取得到的故障暂态波形图像特征和天气特征进行特征融合，实现多模态输电线路故障辨识模型的训练与测试。最后，完全采用真实故障录波数据开展了算例验证。结果表明：所提方法对故障原因的辨识准确率达到了94.87%。相比于传统的故障辨识方法，网络所需的故障特征量更少、对易混淆故障的辨识效果更好、辨识准确度更高。 |
| 关键词: 输电线路 故障辨识 多模态 残差网络 天气特征 |
| DOI：10.19783/j.cnki.pspc.241729 |
| 投稿时间：2024-12-24修订日期：2025-04-09 |
| 基金项目:国家自然科学基金项目资助(52277079)；重庆市留学人员回国创业创新支持计划项目资助(cx2021036) |
|
| Transmission line fault cause identification method based on multimodal residual network integrating waveform and weather information |
| LIN Fengkai1,WANG Jian1,ZHAO Qi2,XUE Han1,PENG Yinzhang2,NAN Dongliang2 |
| (1. State Key Laboratory of Power Transmission Equipment Technology, Chongqing University, Chongqing 400044, China;
2. Electric Power Research Institute of State Grid Xinjiang Electric Power Co., Ltd., Urumqi 830011, China) |
| Abstract: |
| To address the limitation of existing transient waveform image-based transmission line fault cause identification methods, namely, that the use of single-type input features prevents fine-grained fault cause classification, this paper proposes a novel fault cause identification method based on multimodal residual network (ResNet). The method integrates transient waveform features with weather characteristics. First, the characteristics of different causes of transmission line faults are analyzed statistically in terms of both transient waveform and weather conditions. Second, transient waveform images and one-hot codes for weather conditions at the time of the fault are used as inputs to an improved multimodal ResNet classifier. A channel attention mechanism is used to fuse the extracted fault transient waveform image features and weather features, enabling training and testing of the fault identification model. Finally, real fault recording data are used to perform case study verification. The results show that the proposed method achieves a fault cause identification accuracy of 94.87%. Compared with traditional fault identification methods, it requires fewer fault features, offers superior discrimination for easily confusable fault types, and provides significantly higher identification accuracy. |
| Key words: transmission line fault identification multimodal residual network (ResNet) weather feature |
