Research And Application Of GPU-parallel Computing In LSSVM Modeling

Parallel computing is an effective method for large-scale data calculation. Distributed cluster and Super computer are two ways to realize the parallel computing. However, they are usually expensive and strong dedicated, resulting in poor application. Graphic Processing Unit (GPU) is developing rapidly in recent years, whose parallel computing features are determined by its own data-intensive computing model and its own architectures. Therefore, GPU-based graphics and general-purpose computing become a hot research topic in the fields of graphic and high performance computing.The research object of this study is the large-scale and high-performance parallel computing based on GPU, which solves the high-performance computing problem with GPU. On the basis of the implementation of the basic parallel computing, this paper focuses on the parallel method of the least square support vector machine (LSSVM) model based on machine learning. The simulation with real production data shows that the proposed method can greatly improve the efficiency of the learning model. Aiming at the real-time prediction for by-product gas system in steel enterprises, a regression prediction model is established based on the least square support vector machine and the model parameters are optimized. This realization of the gas system prediction and optimization of prediction model parameters on-line are realized by the parallel computing based on GPU, improving the accuracy of the predictions. The simulation results using the practical gas data in Shanghai Baosteel show that the proposed model has better performance. And, in view of the GPU's parallel computing speed and efficiency, the parallel computation speed compared to the traditional serial computing is up to more than 70 times, meeting the requirements of real-time online prediction. Therefore GPU-based parallel computing can be used in practical production.