基于支持向量机回归的城市PM2.5浓度预测

摘要

To establish fast and precise model for prediction of PM 2.5 ,support vector regression (SVR) was used .The concen‐trations of varied air pollutants and meteorological data were trained for building prediction model .Cross‐validation was used for selection of optimal parameters and time window for training data .Optimal SVR model was established and tested on data of several selected cities .The experimental results demonstrate that the SVR method has universality as well as practical value . The proposed method is capable of adjusting optimal parameter for machine learning and enhancing the precision of prediction , compared with other machine learning approached .It provides a facile and efficient method for prediction PM 2.5 concentration in China .%为建立快速精确的PM2．5浓度预测模型，提出利用支持向量机回归（support vector regression ，SVR）方法来建立PM 2.5浓度预测模型。选取各大气污染物浓度以及各气象因素进行训练，对训练好的数据进行交叉验证，取得最优参数和最佳预测特征时间跨度，建立最优PM 2.5浓度的预测模型。基于5个城市的实验结果表明，该方法具有普适性及实际应用意义，能够自适应地调整机器学习最佳参数，相比其它机器学习方法获得了更高的预测精度，为 PM 2.5浓度预测提供了一个简便而有效方法模型。