【摘要】 为了应对城市交通噪声地图的频繁更新，实现大规模环境噪声地图的快速求解，提出了CPU-DCU并行计算方法，并使用“曙光”超级计算平台上的DCU加速卡进行内核计算。首先，设计了噪声计算数据文件，将GIS数据、噪声监测数据等多源异构数据储存为统一的噪声计算数据，通过定向包围盒等方法将建筑物进行几何简化，实现DCU算法优化。其次，针对“曙光”超级计算平台多DCU的优势，实现了在4张DCU加速卡上灵活分配计算任务，4块DCU加速卡对比单DCU加速卡并行效率达到88.2%。最后，对上述方法进行了正确性验证与性能测试，对比了不同计算规模下CPU与DCU的计算效率。结果表明，CPU-DCU并行计算方法能够应用在大规模环境噪声地图研究中，也验证了噪声地图求解在“曙光”超级计算平台系统上的可能性。更多还原

【Abstract】 In order to deal with the frequent updating of urban traffic noise maps and realize the rapid solution of large-scale environmental noise maps, a parallel computing method of CPU-DCU is proposed, and the DCU accelerator card on the Sygon supercomputing platform is used for kernel computing. Firstly, the noise calculation data file is designed, and the GIS data, noise monitoring data and other multi-source heterogeneous data are stored as unified noise calculation data. The buildings are geometrically simplified by means of directional bounding box and other methods to optimize the DCU algorithm. Secondly, in view of the advantages of Sygon supercomputing platform with multiple DCUs, flexible allocation of computing tasks is realized on four DCU accelerator cards, and the parallel efficiency of four DCU accelerators is 88.2% higher than that of a single DCU accelerator. Finally, the correctness and performance of the above methods are verified, and the computing efficiency of CPU and DCU under different computing scales is compared. The results show that the CPU-DCU parallel computing method can be applied to large-scale environmental noise map research, and the possibility of noise map solution on the Sygon supercomputing platform system could be verified.更多还原