含CLCC的变电设备多模态视觉融合感知技术研究
颜楠楠，毛颖科，朱正一，王媚
(国网上海市电力公司，上海 200122)
摘 要 ：具备可控关断能力的换流器 (CLCC) 因其运行工况复杂、瞬态突变频繁且对环境敏感，给传统运维手段带来挑战。提出一种融合视觉基础模型的多模态-多参量感知建模与视觉融合方法，用于设备运行状态的实时识别与三维重构。构建了两阶段三坐标映射模型，融合可见光、红外与声像多源传感器信息。采用平面模板几何标定与奇异值分解 (SVD) 算法实现从相机坐标到世界坐标的几何配准，再经透视投影统一映射至像素平面。为了提高配准精度，引入目标检测模型 YOLOX 和分割模型 SlimSAM 构建辅助感知机制 ：YOLOX 负责生成边界框提示，SlimSAM 通过修剪实现实例级分割与像素级对齐。实验样本覆盖含 CLCC 在内的 11 种典型变电一次设备，实验结果表明 ：CLCC 检测精度较原始 YOLOX 模型提升12.9%，剪枝优化后的 SlimSAM 模型在多类设备实例分割中保持良好的范化性能；在 CLCC 阀厅环境测试中，模型重构耗时低于 68 ms，正确识别率 83.3%，证明该方法在边缘端实现了兼具高精度与快速响应的状态感知能力，对多模态智能运维具有重要应用价值。
关键词 ： 可控关断能力的换流器 ；变电一次设备 ；多模态感知 ；YOLOX 模型 ；SlimSAM 模型
中图分类号 ：TM76 ；TP391.41 文献标识码 ：A 文章编号 ：2097-6623(2026)08-0001-09
Research on Multimodal Visual Fusion Perception Technology for
Substation Equipment Including CLCC
YAN Nan-nan, MAO Ying-ke, ZHU Zheng-yi, WANG Mei
（ State Grid Shanghai Municipal Electric Power Company, Shanghai 200122, China）
Abstract: Conventional operation and maintenance methods are challenged by the controllable line commutated converter (CLCC) due to its complex operating conditions, frequent transient state mutations and environmental sensitivity. To this end, this paper proposes a multimodal, multi-physical perception modeling and visual fusion method that integrates a vision foundation model, aiming at real-time identification and three-dimensional reconstruction of equipment operating states. A two-stage three-coordinate mapping model is constructed to fuse multi-sensor information from visible light, infrared, and acoustic imaging. Geometric calibration from camera coordinates to world coordinates is achieved through planar template-based geometric calibration and singular value decomposition (SVD), followed by a unified mapping to the pixel plane via perspective projection. To enhance registration accuracy, an auxiliary perception mechanism is introduced by incorporating the YOLOX object detector and the SlimSAM segmentation model: YOLOX generates bounding-box prompts, while SlimSAM, with its pruned architecture, enables instance-level segmentation and pixel-wise alignment. Experimental samples cover 11 typical types of primary substation equipment, including the CLCC. The results demonstrate that CLCC detection accuracy improves by 12.9% over the baseline YOLOX model, and the pruned SlimSAM maintains robust generalization performance across multiple equipment categories. In the CLCC valve hall environment, the proposed method achieves a reconstruction time below 68 ms and a correct recognition rate of 83.3%, confirming that the approach enables both high-precision and rapid-response condition awareness at the edge side, offering significant application value for multimodal intelligent operation and maintenance.
Key words: controllable line commutated converter; primary substation equipment; multimodal perception; YOLOX model; SlimSAM model
参考文献
[1] 张静，高冲，杨俊，等 . 可控换相换流器耗能模式工作原理及电气建模 [J]. 高电压技术，2024，50(7) ：2784-2793.
[2] 高冲，贺之渊，杨俊，等 . 新型可控电网换相换流器拓扑及其控制方法 [J]. 中国电机工程学报，2023，43(5) ：1940-1949.
[3] 蒲天骄，乔骥，韩笑，等 . 人工智能技术在电力设备运维检修中的研究及应用 [J]. 高电压技术，2020，46(2) ：369-383.
[4] 李登攀，任晓明，颜楠楠 . 基于无人机航拍的绝缘子掉串实时检测研究 [J]. 上海交通大学学报，2022，56(8) ：994-1003.
[5] Jiang Anfeng， Yan Nannan， Shen Baoguo， et al.Research on infrared image recognition method of power equipment based on deep learning[C]//IEEE International Conference on High Voltage Engineering and Application(ICHVE)，2020 ：1-4.
[6] 李兵，邓慰，鄢哲，等 . 基于紫外成像的电晕放电环境影响因素的研究 [J]. 高压电器，2015，51(1) ：24-25.
[7] Yan Nannan， Zhu Huangru， Qian Chengchen， et al.Analysis of multi-spectral signal in GIS and comparing with electrical signals[C]//International Conference on Electrical Materials and Power Equipment(ICEMPE)，2021 ：1-4.
[8] Li Zhihe， Wang Youyuan， Zhang Zhanxi， et al.Localization of mechanical and electrical defects indry-type transformers using an optimized acoustic imaging approach[J].PLoS ONE，2023，18(11) ：e0294674.
[9] Caron M， Touvron H， Misra I， et al.Emerging properties in self-supervised vision transformers[C]//IEEE/CVF International Conference on Computer Vision(ICCV)，2021 ：9650-9660.
[10] Yang Lihe， Kang Bingyi， Huang Zilong， et al.Depth anything ：unleashing the power of largescaleunlabeled data[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR)，
2024 ：10371-10381.
[11] Kirillov A， Mintun E， Ravi N， et al.Segment anything[C]//IEEE/CVF International Conference on Computer Vision(ICCV)，2023 ：4015-4026.
[12] Wang Yulin， Huang Rui， Song Shiji， et al.Not all images are worth 16×16 words ：dynamic transformers for efficient image recognition[J].Advances in Neural Information Processing Systems，2021，34 ：11960-11973.
[13] Cheng Bowen， Misra I， Schwing A G， et al.Maskedattention mask transformer for universal image segmentation[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR)，2022 ：1290-1299.
[14] Cheng Bowen， Schwing A， Kirillov A.Per-pixel classification is not all you need for semantic segmentation[J].Advances in Neural Information Processing Systems，2021，34 ：17864-17875.
[15] Ravi Nikhila， Gabeur Valentin， Hu Yuanting，et al.SAM 2 ：segment anything in images and videos[PP/OL].V2.arXiv(2024-10-28)[2025-10-22].https ：//doi.org/10.48550/arXiv.2408.00714.
[16] Zhang Chaoning， Han Dongshen， Qiao Yu，et al. Faster segment anything ：towards lightweight SAM for mobile applications[PP/OL].V2.arXiv(2023-07-01)[2025-10-22].https ：//doi.org/10.48550/arXiv.2306.14289.
[17] Shu Han， Li Wenshuo， Tang Yehui， et al.TinySAM ：pushing the envelope for efficient segment anything model[PP/OL].V3.arXiv(2025-01-08)[2025-10-22].https ：//arxiv.org/pdf/2312.13789v3.
[18] Touvron H， Cord M， Douze M， et al.Training dataefficient image transformers & distillation through attention[C]//International Conference on Machine Learning， 2021 ：10347-10357.
[19] Wu Kan， Zhang Jinnian， Peng Houwen， et al.TinyVit ：fast pretraining distillation for small vision transformers[C]//European Conference on Computer Vision(ECCV)， 2022 ：68-85.
[20] Chen Zigeng， Fang Gongfan， Ma Xinyin， et al.SlimSAM ：0.1% data makes segment anything slim[PP/OL].V3.arXiv(2024-09-26)[2025-10-22].https ：//doi.org/10.48550/arXiv.2312.05284.
[21] Zhou Chong， Li Xiangtai， Loy Chen Change， et al.EdgeSAM ：prompt-in-the-loop distillation for on-device deployment of SAM[PP/OL].V3 arXiv(2025-09-07)[2025-10-22].https ：//doi.org/10.48550/arXiv.2312.06660.
[22] Lin Tsung-Yi， Maire Michael， Belongie Serge， et al.Microsoft COCO ：common objects in context[C]//European Conference on Computer Vision(ECCV)，2014 ：740-755.
[23] Jocher G，Chaurasia A，Stoken A，et al.Ultralytics/yolov5：v7.0-YOLOv5 SOTA realtime instance segmentation[DB/OL].Zenodo(2022)[2025-10-22].https ：//ui. adsabs.harvard.edu/abs/2022zndo...7347926J/metrics.DOI ：10.5281/zenodo.7347926.
[24] Lin Tsung-Yi， Dollár Piotr， Girshick Ross，et al.Feature pyramid networks for object detection[C]//IEEE Conference on Computer Visionand Pattern Recognition(CVPR)，2017 ：2117-2125.
[25] Neubeck Alexander， Van Gool Luc.Efficient nonmaximum suppression[C]//18th International Conference on Pattern Recognition(ICPR’06)，2006 ：850-855.
[26] Chen Kai， Wang Jiaqi， Pang Jiangmiao， et al.MMDetection ：open MMlab detection toolbox and benchmark[PP/OL].(2019-06-17)[2025-10-22].https ：//arxiv.org/pdf/1906.07155.
[27] He Kaiming， Zhang Xiangyu， Ren Shaoqing， et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2015，37(9) ：1904-1916.
[28] Liu Shu，Qi Lu， Qin Haifang， et al.Path aggregation network for instance segmentation[J].IEEE/CVF Conference on Computer Vision and Pattern Recognition，2018 ：8759-8768.
[29] Ramachandran Prajit， Zoph Barret，Le Quoc V.Searching foractivation functions[PP/OL].V2.arXiv(2017-10-27)[2025-10-22].https ：//arxiv.org/pdf/1710.05941.
