基于并行SIMT平台的电网系统造价数据分析研究  

Analysis and study of power grid system cost data based on parallel SIMT platform

作　　者：郭嘉成 宋妙环 王炳文 GUO Jiacheng;SONG Miaohuan;WANG Bingwen(Economic and Technological Research Institute,State Grid Gansu Electric Power Company,Lanzhou 730050,China)

机构地区：[1]国网甘肃省电力公司经济技术研究院,甘肃兰州730050

出　　处：《电子设计工程》2021年第20期129-133,共5页Electronic Design Engineering

基　　金：甘肃省电力公司电力经济技术研究院科技项目(5227-2817-000R)。

摘　　要：电网工程造价数据庞杂、分析难度高,而传统基于CPU的数据分析方法效率与精度均较低。针对此问题,文中使用SIMT CPU平台设计了一款高效、迅速的电网工程造价数据分析系统。针对电网数据的特点,分别设计了GPU加速混合多网格算法、GPU HMD算法及工作负载分区策略。为解决运算速度与内存访问之间的平衡问题,文中通过减少随机内存访问模式简化了控制,有效提高了计算效率。实验结果表明,文中所提数据分析系统可以在直接求解器上实现100倍的运行时加速,且较基于CPU的多网格分析方案速度提高了15倍以上。The power grid project cost data is complex and difficult to analyze,while the traditional CPU-based data analysis method has low efficiency and accuracy.In response to this problem,this paper uses the SIMT CPU platform to design an efficient and rapid power grid engineering cost data analysis system.For the characteristics of grid data,GPU-accelerated hybrid multi-grid algorithm,GPU HMD algorithm and workload partition strategy are designed respectively.In order to solve the problem of the balance between computing speed and memory access,the article simplifies the control by reducing the random memory access mode and effectively improves the computational efficiency.The experimental results show that the data analysis system proposed in the article can achieve a run-time acceleration of 100 times on the direct solver,and is more than 15 times faster than the CPU-based multi-grid analysis scheme.

关 键 词：图形处理器 SIMT 电网工程 数据分析 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]