Research Of Support Vector Regression Modeling Based On Spatial Analysis And Parallel Computing

Spatial analysis refers to the description and analysis of spatial phenomena from the perspective of geography,which aims to study of the location and properties of spatial target through geographical calculation and spatial expression to mine the potential spatial information.However,with the advent of the era of big data,spatial analysis is faced with increasingly complex objects and massive information,traditional spatial analysis methods can't meet the requirements of geographic information system and geo-information science processing,which makes the Artificial Neural Network(ANN),Support Vector Machine(SVM)and other intelligent computing technology develop rapidly.At the same time,the real-time processing of massive data makes it difficult to rely solely on the Central Processing Unit(CPU)and can't meet the needs of geospatial analysis.The key to solve the problem is combined with Graphics Processors Unit(GPU)for numerical parallelization operations.Spatial analysis,parallel computing technology,intelligent computing and other different theoretical methods of information fusion technology is the trend of the times.In this paper,the main contents are as follows:(1)Based on the Kriging interpolation model,the multi-scale wavelet least squares support vector regression model is introduced to optimize the conventional kriging interpolation method,which makes the optimized method more accurately estimate the position site of the attribute information and is applied to the concentration of Qingdao PM2.5 interpolation.(2)According to the spatial dependence analysis of Tobler's First Law of Geography,the form of spatial weight matrix is introduced as the measure of the interaction between spatial objects.The least squares support vector regression fusion model of spatial dependence is established to realize the spatial data regression analysis and performance evaluation by combining with multiple test data.(3)Based on NVIDIA's Compute Unified Device Architecture(CUDA)parallel architecture,the experimental platform of CUDA,Visual Studio and Matlab are established and the parallel computation of kernel function matrix are completed.At last,we achieve the acceleration of least squares support vector regression fusion model of spatial dependence.This paper realizes the support vector regression model based on spatial analysis and the parallel computing research,which not only improves the computational efficiency,but also improves the ability of geospatial data to explain the geospatial phenomena.