http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
A GPU-based Two Dimensional Flood Simulation System
Kuo-Chan Hung,Chih-Yung Feng,Kun-Huai Cheng,Pei-Cheng Tsao1,Hung-Kwai Chen 한국방재학회 2016 Journal of Disaster Management Vol.1 No.1
Abstract A GPU-based flood simulation system is developed in this research. The system comprises two main components: a GPU-lized two bdimensional flood simulation model named GLMRT and a web-GIS user interface. GLMRT is designed and constructed with following features. First, GLMRT is constructed based on Lattice Boltzmann method(LBM) and two dimensional shallow water equations. Second, GLMRT is combined with sink/source terms including excess rainfall and pump-like drainage. Third, a dry-wet boundary determining mechanism is considered in GLMRT. Fourth, real DEM data can be used in simulating water movement on the land surface. Finally, GPU accelerated technique is integrated into GLMRT to overcome the issue of time cost in calculation. The web-GIS based user interface is built based on cloud platform to assist data flow organization and real time simulation. The interface integrates real DEM data and receives precipitation estimation products from the Taiwan Typhoon and Flood Research Institute(TTFRI) and Central Weather Bureau(CWB). The simulation results would be displayed on the web-GIS via file and database transfer. Besides, a real time rendering technique based on real time simulation is developed, too. The real time rendering technique displays the simulation results via memory on GPU device and eliminates the time cost in data output. GLMRT solves a 800-second simulation of Zengwen Reservoir virtual dam break with 2,046x1,525 DEM grids (40 m resolution) within 140.11 seconds on the NVIDIA Tesla K20c. Compare to the 60,919 seconds computation time on the Intel?? Xeon?? CPU X5650 2.66GHz, the GLMRT gets a speedup of 438x. On the other hand, using GLMRT alone to simulate a 12-hour rainfall duration induced inundation would spend about 1.2 hours, which can provide buffer time for the requirement of flood prevention operation. Abstract A GPU-based flood simulation system is developed in this research. The system comprises two main components: a GPU-lized two bdimensional flood simulation model named GLMRT and a web-GIS user interface. GLMRT is designed and constructed with following features. First, GLMRT is constructed based on Lattice Boltzmann method(LBM) and two dimensional shallow water equations. Second, GLMRT is combined with sink/source terms including excess rainfall and pump-like drainage. Third, a dry-wet boundary determining mechanism is considered in GLMRT. Fourth, real DEM data can be used in simulating water movement on the land surface. Finally, GPU accelerated technique is integrated into GLMRT to overcome the issue of time cost in calculation. The web-GIS based user interface is built based on cloud platform to assist data flow organization and real time simulation. The interface integrates real DEM data and receives precipitation estimation products from the Taiwan Typhoon and Flood Research Institute(TTFRI) and Central Weather Bureau(CWB). The simulation results would be displayed on the web-GIS via file and database transfer. Besides, a real time rendering technique based on real time simulation is developed, too. The real time rendering technique displays the simulation results via memory on GPU device and eliminates the time cost in data output. GLMRT solves a 800-second simulation of Zengwen Reservoir virtual dam break with 2,046x1,525 DEM grids (40 m resolution) within 140.11 seconds on the NVIDIA Tesla K20c. Compare to the 60,919 seconds computation time on the Intel?? Xeon?? CPU X5650 2.66GHz, the GLMRT gets a speedup of 438x. On the other hand, using GLMRT alone to simulate a 12-hour rainfall duration induced inundation would spend about 1.2 hours, which can provide buffer time for the requirement of flood prevention operation.