The Research On Parallel Computing Technology In Precise Agricultural Climate Division

With the changes of global climate conditions, as well as the rapid development ofmodern agricultural production, the analysis and application of agricultural climateresources are putted forward higher requirements. The core technology of the modernagricultural climatic division contains small grid interpolation techniques andagricultural climate resources division.Small grid interpolation technique has been widely used in the analysis of climateresources. It is based on geographic information system technology. And it is animportant technique for the research of the information of various application areas.Meanwhile, the small grid interpolation technique processes mass data. It requires alot of computing time and can not meet the requirements of real-time analysis. Inorder to improve the response time of system and the efficiency of informationprocessing, we use Data parallel strategy and master/slave programming model todevelop the parallel algorithm of Kriging. We also take advantage of dynamic loadbalancing to further improve the efficiency of the algorithm.The techniques using in zoning of agricultural climate resources are clusteringanalysis, weight method, and expert scoring method and so on. K-means is a wellknown partitioning clustering technique which uses deviation as clustering criteria.And it is widely used because its simplicity and speed are very appealing in practice.Precise agricultural climatic division makes use of K-means clustering algorithm inagricultural climate resources division. For huge volume of the data of small gridinterpolation and the need for real-time clustering, the classical K-means need to beimproved. In order to make better accuracy and efficiency of traditional K-meansclustering algorithm, we proposed two new algorithms: the CK-means algorithm andthe K-means#algorithm. CK-means is an improved algorithm based on competitivestrategy, which aims at the small grid interpolation data (large amount of onedimensional data). It has the same regionalization of climate resources as K-means,but it is superior in operating efficiency. At present, CK-means has been successfullyapplied to fine agricultural climate division. We take advantage of the "D2seeding"method from K-means++, and propose an improved algorithm called K-means#whichintroduces the concept of "Neighbor set of cluster centers". Both theoretical analysisand experimental results show that, K-means#not only improves the accuracy of K-means, but also significantly accelerates the operating efficiency.To further improve the efficiency of cluster analysis, the paper also discusses theparallel K-means++algorithm as well as the parallel K-means#algorithm.Considering the characteristics of K-means++and the capabilities of the machine node,the use of data parallel and load balancing is adopted again. The results showed that:the parallel K-means++algorithm has the same clustering result as that of the serial,but the efficiency has been greatly improved. For K-means#, we also give a viableparallel idea and list the algorithm process, which provides a meaningful reference forthe further research.