Abstract: To address the challenges of low data transmission efficiency, high storage overhead, and limited rendering frame rates of applying 3D Gaussian Splatting to large-scale power grid scenes, we propose a lightweight cross-platform rendering system to support the efficient construction of smart grid digital twins. The sys-tem comprises the following components: First, Gaussian primitive attributes are quantized and encoded, and images are compressed using the WebP format to reduce storage footprint. Second, the scene is spa-tially partitioned using an octree structure to establish a multi-level hierarchical data organization. Third, a dynamic LOD scheduling strategy is designed based on viewpoint distance and view frustum range, ena-bling on-demand loading of Gaussian primitive data at different detail levels. Finally, all modules are inte-grated into a cross-platform rendering pipeline to support deployment and real-time rendering on both PC and Web clients. Experimental results on three UAV-captured power grid scenes and three outdoor re-al-world datasets demonstrate that the proposed system achieves a data compression ratio of 19.46% and an average Web-client rendering frame rate of 58 frames/s, significantly reducing storage overhead and improving transmission efficiency while maintaining acceptable visual quality.