| Kriging algorithms are a group of important interpolation methods, which are very usefulin many engineering applications. However, the algorithm based on traditional general purposeprocessors can be computationally prohibitive as the problem scale expands. Therefore, it isurgent to design a reasonable, effective and reusable parallel Kriging algorithm.This dissertation mainly devotes to use GPGPU architecture by some of the newestresearch advanced results, with component oriented design related theory, to solve the problemin the process of performance bottlenecks. Around the research orientation, the work of thispaper content and main contribution as follows:1) Study all the nuclear system structure characteristics and CUDA parallel programmingmodel, the main body of Kriging algorithm computing tasks, and improve theperformance of the restricted factors were analyzed. Based on this, we put forwardbased on NVIDIA CUDA platform of a general Kriging algorithm is effective parallelstrategies.2) In the numerical experiment, the experimental data is come form a hillside data whichis observed through a geeological wb site. We guarantee the correctness of the data inthe premise, to compare the different versions of the algorithm to optimize the CPUand GPU on the performance of the operation. After optimization based on CUDArealize interpolation algorithm and the original serial algorithm, have the goodperformance of the calculation, accelerate than to18. The practical application in theexperiment, first of180layers of groundwater model data preprocessing ordos, thenthe interpolation, original point for469440, after the number of the interpolation pointsfor1877760, accelerate topped than38, which could verify this topic research has highpractical value.3) Use technology to parallel component oriented to encapsulate Kriging algorithm toshield the complexity of the parallel algorithm, encapsulation algorithm of its goodscalability and portability. |
PDF Full Text Request