摘要
利用计算网格实现高效率、低误差的时间序列预测,对科研、工商业等各个领域都具有重要的现实意义。使用Nu-支持向量回归方法建模时间序列预测问题;提出了数据集预处理方法,将原始时间序列转换成标准化的标记样本集;为了优化预测模型的参数,基于并行化和粒度控制提出两阶段搜索策略。在网格计算环境内设计了系统框架,以网格服务的动态组合实现时间序列预测。使用基准数据集对系统化预测方案进行性能测试,优化结果表明本方案能够针对特定数据集自适应的完成模型参数优化,且显著加速了优化过程。预测结果显示优化后的模型针对未知样本能获得较高的预测精度。
Appling Computing Grid to achieve time series prediction with high efficiency and low error has significant meaning to most practical fields such as scientific research, industry, commerce and so on. In this paper, Nu-support vector regression is employed as method to model time series prediction; data set preprocessing method is proposed to transform original time series to standardized set of labeled samples ; in order to optimize parameters of prediction model, twostage search strategy is proposed based on parallelization and granularity control. System framework is designed under Computing Grid environment to achieve time series prediction based on dynamically collaboration of Grid services. Benchmark data set is used to test the performance of systematic prediction solution proposed. Optimizing results indicate that our sol- ution is capable of optimizing model parameters according to individual data set adaptively, and the optimizing process is remarkably accelerated. Prediction results show that the optimized model can keep up with high accuracy for unseen samples.
出处
《长春理工大学学报（自然科学版）》
2009年第2期273-278,共6页
Journal of Changchun University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金项目(60873235
60473099)
吉林省科技发展计划重点项目(20080318)
教育部新世纪优秀人才基金项目(NCET-06-0300)
关键词
Nu-支持向量回归
时间序列预测
网格服务
参数优化
粒度控制
Nu-support vector regression
time series prediction
Grid service
parameter optimization
granularity control
