【摘要】 针对空气污染物浓度变化的复杂性、非线性等特点，建立机器学习模型对污染物浓度变化预测。通过分析呼和浩特市不同时间尺度下空气污染物时间变化特征，确定把空气污染浓度、气象因素、季节因素和人为因素作为预测模型输入变量。利用灰色关联度分析确定与PM_2.5和PM₁₀关联度高的空气污染物，利用主成分分析法提取气象因素中主要气象因子，建立BP神经网络模型和支持向量机模型用于预测PM_2.5和PM₁₀的浓度。BP神经网络模型的预测值与实测值接近，训练集和测试集的R²均达到0.8以上；平均绝对误差均小于19；均方根误差均小于25。SVM模型对PM_2.5和PM₁₀预测的R²分别为0.895和0.668。更多还原

【Abstract】 A machine learning model was established to predict the change of pollutant concentration according to the complexity and nonlinearity of air pollutant concentration change.By analyzing the time variation characteristics of air pollutants at different time scales in Hohhot,the air pollution concentration,meteorological factors,seasonal factors and human factors were determined as the input variables of the prediction model.The air pollutants with high correlation with PM_2.5and PM₁₀were determined by grey correlation analysis.The main meteorological factors in meteorological factors were extracted by principal component analysis.BP neural network model and support vector machine model were established to predict the concentration of PM_2.5and PM₁₀.The predicted value of BP neural network model is close to the measured value,and the R²of training set and test set is above 0.8.The mean absolute error （MAE） was less than 19.The root mean square error（RMSE） was less than 25.The R²of the SVM model for PM_2.5prediction can reach 0.895,but the prediction R²for PM₁₀is only 0.668.更多还原