| Along with the proposition of Action Plan for "Prevention and Control of Water Pollution", "Ecological Environment Monitoring Network Construction Plan" and more and more policies about environmental protection and environmental monitoring scheme, China has entered into an overall land-water coupled and river-ocean combined water environmental management era. Based on big-data and power-efficient computer systems, the over-all environmental analysis technique and model methods are important issues which currently need breakthrough in its research and implementation. In consideration of problems of overall environmental management in China, this paper constructs a highly productive cluster computing system, for large-scale basin environmental simulation. Based on big-data theory as well as parallelism and clustering techniques, the model can reach the goal of high accuracy, adjustable scaling and lower cost computing. The main contents are as followings:(1) Based on the mixed MPI+OPENMP computing technology, aimed at the characteristics of the large-scale watershed and massive spatial computing unit and grid, to face the service concept of environmental big data and management business, an three degree coupling parallel cluster technology among models, computing nodes and modules had been proposed creatively. And the watershed environment high performance cluster computing model implemented the reliability of precision, available validity, adjustable scale and low(2) Aimed at the real-time, complexity and mass of environmental big data, watershed environment model cluster management system which adapted to the environment data characteristics and serviced in the environment management business was built. And it concluded data centers, model centers, management centers and business centers. Under the dispatch of management centers, data centers provided data for model centers, and model centers serviced for requirements and data characteristics of business centers.(3) Breaking through the three-dimensional hydrodynamic and water quality model of huge grid construction technology, based on the partition technology of watershed environment model with overlap region data exchange, and the watershed environment model grid achieved the spliced, extensible and separable and built a balanced blocking. It provided security to the reality of the parallel computing grid data of large-scale watershed model.(4) This paper choose the principal river of Yangtze river of 72km length, Zhuxi river section of 8km length and Pengxi river section of 21km, also with their confluence area of 6639.79km2 from The three Gorges Reservoir as demonstration sites. Those demonstration sites were divided into 385 subbasins,850 hydrological response units,719141 grids and 14 nodes, to testify the application of the clustering models. The results show the linear regression coefficient R2 and the Nash-Sutcliffe Ens of the contaminant transport model were more than 0.75, the simulated error of hydrodynamic model was less than 10%. The simulation data can joint very well with practical ones. It proves that these models are more accurate and proficient based on parallelism and clustering techniques.(5) By setting different schemes of cluster computing, the high-performance cluster computing model was tested, and the impact of data exchange, network communications for cluster computing was evaluated. The results showed an exponential relationship between the regional block numbers and the calculation time. The data exchange and network latency would reduce the efficiency of the whole cluster computing. The cluster parallel and serializing computing for the same area was studied. Results shows system lifetime of cluster computing is improved by 7.92%, and the parallel computing is improved by 36.01%. The study area was divided into 200 cluster computing nodes, comparing with 14 nodes computing, it needed two days and 1 h54min to simulate and 98% increase in efficiency. The cluster computing model can satisfy business management needs of time and effectiveness in large scale environment. |
PDF Full Text Request