Scholarlink

<p>This study addresses the critical challenges in the 3D reconstruction of power grid equipment, namely, geometric-semantic fragmentation, low dynamic updating efficiency, and the trade-off between high precision and lightweight representation. We propose a novel method that integrates Neural Radiance Fields (NeRF) with lightweight Building Information Modeling (BIM). Our core contributions are as follows. First, a cross-modal alignment mechanism featuring an Attribute-Aware Attention (AAA) Module dynamically embeds BIM semantics into the feature space of NeRF, achieving a unified representation. Second, an incremental updating algorithm, triggered by Structural Similarity Index (SSIM) difference analysis and BIM change logs, employs the Adaptive Boundary (AdaBound) optimizer for local fine-tuning and a Git-style storage strategy, reducing single-device update latency to 3.2&#xa0;min and storage overhead by 98%. Third, a cloud-edge-end collaborative architecture leverages knowledge distillation to compress the model to 0.3&#xa0;M parameters and a static-dynamic hierarchical rendering strategy to enable real-time mobile applications. Experiments in a 220&#xa0;kV substation demonstrated the superiority of our method over traditional BIM and standard NeRF, with a Chamfer Distance of 2.03&#xa0;cm, semantic recall of 98.7%, and a 98% transmission success rate under weak network conditions. This study provides key technical support for smart grid inspection and dynamic operational analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

3D reconstruction and dynamic updating of power grid equipment via fusion of NeRF and lightweight BIM

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月21日

查看全文
  • Ming Zhang,
  • Beibei Li,
  • Shuhui Yu,
  • Xiaoyin Sun

摘要

This study addresses the critical challenges in the 3D reconstruction of power grid equipment, namely, geometric-semantic fragmentation, low dynamic updating efficiency, and the trade-off between high precision and lightweight representation. We propose a novel method that integrates Neural Radiance Fields (NeRF) with lightweight Building Information Modeling (BIM). Our core contributions are as follows. First, a cross-modal alignment mechanism featuring an Attribute-Aware Attention (AAA) Module dynamically embeds BIM semantics into the feature space of NeRF, achieving a unified representation. Second, an incremental updating algorithm, triggered by Structural Similarity Index (SSIM) difference analysis and BIM change logs, employs the Adaptive Boundary (AdaBound) optimizer for local fine-tuning and a Git-style storage strategy, reducing single-device update latency to 3.2 min and storage overhead by 98%. Third, a cloud-edge-end collaborative architecture leverages knowledge distillation to compress the model to 0.3 M parameters and a static-dynamic hierarchical rendering strategy to enable real-time mobile applications. Experiments in a 220 kV substation demonstrated the superiority of our method over traditional BIM and standard NeRF, with a Chamfer Distance of 2.03 cm, semantic recall of 98.7%, and a 98% transmission success rate under weak network conditions. This study provides key technical support for smart grid inspection and dynamic operational analysis.