Refined 3D Gaussian Splatting Method for Distribution Network Modeling Towards AR-Oriented Operation and Maintenance

Aiming at the problems of missing slender structure modeling, complex background interference, and algorithm efficiency bottlenecks in the 3D reconstruction of distribution networks, this paper proposes an improved 3D Gaussian splatting-based algorithm for 3D reconstruction of distribution network scenes. First, through a semantics-guided density control strategy combined with adversarial background suppression, power equipment and vegetation artifacts are effectively separated. Second, a direction-sensitive covariance optimization method is designed to enhance the geometric continuity and detail fidelity of ultra-slender structures such as power lines and utility poles. Finally, a lightweight progressive splatting framework is constructed to achieve real-time rendering at ≥35 fps under 1080p resolution. Experiments show that the improved algorithm significantly increases the completeness rates of power lines, utility poles, and insulators to 90%, 95%, and 80%, respectively, reduces background artifacts to 2% in area, and shortens the reconstruction time to 0.4 hours. This study provides a high-precision modeling tool for digital twins of distribution networks and supports real-time applications of augmented reality (AR) technology in inspection navigation, fault localization, and remote collaboration.