基于分布式并行计算架构的电力工程三维地理信息产品高效生产应用研究

Research on Efficient Production Application of Three-dimensional Geographic Information Products for Power Engineering Based on Distributed Parallel Computing Architecture

摘要

针对电力工程测量领域三维地理信息产品生产的算力瓶颈，探究分布式并行计算架构的应用效能。以海南光伏送出线路等四项实际工程为载体，选取75-104 GB的无人机影像与激光点云为样本，采用“单机基准+集群对比”方案，从处理耗时、资源利用率及成果精度三方面，对比分析5节点集群与单机处理的差异。结果表明：集群模式下，正射影像处理效率提升247%-263%，空三耗时缩短72%-74%；倾斜摄影建模效率提升250%，多视角空三解算时间压缩74%；点云分类与DTM生成效率提升255%，分类耗时缩短72%。各节点CPU占用率稳定在70%-90%，负载均衡性优于单机（持续95%以上满载）。成果精度（点云分类准确率92%、模型几何中误差±2-3 cm）与单机完全一致。结论：分布式并行计算架构可在保障精度的前提下，将生产周期压缩至传统模式的1/3以下，有效化解算力瓶颈，为电力行业测绘数字化转型提供技术参考。

Abstract

To address the computational bottleneck in the production of three-dimensional geographic information products for power engineering surveying, this study explores the application efficiency of distributed parallel computing architecture. Using four practical projects such as Hainan photovoltaic transmission lines as case studies, and selecting 75-104 GB of drone imagery and laser point clouds as samples, a “single-machine benchmark + cluster comparison” approach was adopted to analyze the differences between a 5-node cluster and single-machine processing in terms of processing time, resource utilization rate, and output accuracy. The results demonstrate that under the cluster mode, orthophoto processing efficiency improves by 247%-263% while Air Triangulation (AT) processing time is reduced by 72%-74%. Oblique photography modeling efficiency increases by 250%, and multi-view AT solution time is compressed by 74%. Point cloud classification and Digital Terrain Model (DTM) generation efficiency rise by 255%, with classification time reduced by 72%. CPU utilization across nodes remains stable at 70%-90%, exhibiting superior load balancing performance compared to single-machine operation (maintaining over 95% full load capacity). The output accuracy (point cloud classification accuracy of 92%, model geometric mean error within ±2-3 cm) is fully consistent with single-machine configurations. Conclusion:The distributed parallel computing architecture can compress production cycles to less than one-third of the traditional model while maintaining accuracy, effectively alleviating computational capacity bottlenecks and providing technical references for the digital transformation of power industry surveying and mapping.

关键词

分布式计算 / 三维地理信息 / 生产效率

Key words

distributed computing / 3D geographic information / production efficiency

参考文献

[1] 国网电力空间技术有限公司. 电网空间数据云上处理平台建设与应用[R]. 北京: 国家电网有限公司, 2025.

[2] 金坚, 陈成, 冯雷, 等. 面向输电线路建设规划的三维可视化交互系统研究[J]. 东北电力技术, 2026, 47(2): 34-41.

[3] 王海波, 张立明, 赵建辉. 基于云边协同的输变电工程建设调度平台设计与应用[J]. 电力信息与通信技术, 2025, 23(4): 56-63.

[4] Chang A, Jung J, Landivar J, et al. Performance evaluation of parallel structure from motion processing with public cloud computing and on-premise cluster system for UAV images in agriculture[J]. ISPRS International Journal of Geo-Information, 2021, 10(10): 677.

[5] 郭建平, 饶晨, 陈永红. 基于云计算的电网三维点云数据分析技术研究与应用[C]// 中国智能电网学术研讨会论文集. 北京: 电子工业出版社, 2023: 95-103.

[6] 李德仁, 王密, 沈欣. 从数字地球到智慧地球[J]. 武汉大学学报(信息科学版), 2023, 48(1): 1-12.

[7] 龚健雅, 季顺平. 摄影测量与深度学习[J]. 测绘学报, 2024, 53(2):203-218.

[8] 杨必胜, 董震. 点云智能处理及其在电力巡检中的应用[J]. 测绘地理信息, 2025, 50(3): 1-10.

[9] 张祖勋, 陶鹏杰. 谈大数据时代的“云控制”摄影测量[J]. 测绘学报, 2024, 53(5): 801-812.