基于主曲线的空气悬浮颗粒物质PM10的预测

The Prediction of Air Suspended Particulate Matter PM 10 Based on the Principal Curve

对空气中有害物质(例如PM10)浓度的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、误差小,同时网络结构更加紧凑.

It＇s of very practical significance to predict the density of hazardous substance（ such as PM10） in the air,but in the overwhelming majority of cases,this kind of data have the characteristics of imbalance and sequential arrived online,and it＇s difficult to realize rapid and effective prediction by traditional supervised learning methods. To solve this problem,this paper proposes a PM 10 prediction method based on the principal curve,building a PM10 model between 2010 and 2012,receiving corresponding parameters by fitting,finally getting the prediction model of the principal curve,and setting corresponding threshold values of different density of PM10 by a lot of experiment. The research shows that the PM10 prediction model based on the principal curve predicts rapidly and has a lower prediction error,meanwhile,the network structure is more compact.