摘要
对空气中有害物质(例如PM10)的浓度进行预测具有重要的现实意义,但绝大多数情况下,这类数据具有不均衡、在线贯序到达的特点,利用传统监督学习方法较难以实现快速、有效的预测。为解决该问题,提出了一种基于主曲线的PM10预测方法,建立在2010年到2012年PM10的模型,拟合得到相应的参数,最终得到主曲线预测模型,并通过大量实验分别设定不同浓度PM10相应的阈值。研究表明,基于主曲线的PM10预测模型预测速度快、误差低,同时网络结构更加紧凑。
It＇s very practical significance to predict the density of hazardous substance（ such as PM10 ） incases, this kind of data have the characteristics of imbalance and sequential arrived online. It’s difficult to realize rapid and effective prediction by traditional supervised learning methods. In order to solve this problem, a PM 10 prediction method which based on the principal curve,build a PM10 model of PM10 from 2010 to 2012, received corresponding parameters by fitting. Finally,the main curve is obtained, and set corresponding threshold values of different density of PM 10 respectively by a lot of experiment. The results showthat the PM10 prediction model based on the principal curve predicts rapidly and a lower prediction error,meanwhile, the net-work structure is more compact.
出处
《南阳理工学院学报》
2017年第2期19-22,共4页
Journal of Nanyang Institute of Technology
基金
国家自然科学基金(U1204609)
关键词
主曲线
悬浮微粒物质PM10
不均衡
the principal curve
suspended particulate matter PM 10 ＆ imbalance
