【摘要】 建立基于LIBSVM方法的重污染天气PM_2.5、PM₁₀预测模型及参数寻优,得到重污染天气时PM_2.5、PM₁₀对应的最优环流形势场并分析重污染天气环流特征。（1）根据与PM_2.5的相关系数的绝对值及显著性,在训练值占比69%,相关性与显著性最好的27个要素参与时得到了重污染天气下PM_2.5日数据最优预测模型,训练值和测试值的R²分别达到了0.9992和0.7196;（2）由相关性与显著性最好的24个要素参与,得到了重污染天气下PM₁₀日数据最优预测模型,训练值和测试值的R²分别达到了0.9978和0.7792;（3）PM_2.5最优预测模型对应的最优5类700 hPa合成环流形势中第4、5类和PM₁₀模型对应的最优6类700 hPa合成环流形势中第1、3、4类均属于高压脊控制（影响）型,容易出现长时间持续污染,是天目山地区影响PM_2.5和PM₁₀污染的最主要天气类型。更多还原

【Abstract】 The LIBSVM-based prediction model and parameter optimization of PM_2.5 and PM₁₀ in heavy pollution weather are established. The optimal circulation fields corresponding to PM_2.5 and PM₁₀ in heavy pollution weather are obtained and the circulation characteristics of heavy pollution weather are analyzed. The results show that:（1）according to the absolute value significance of correlation coefficient with PM_2.5, the optimal prediction model of PM_2.5 data in heavy pollution weather is obtained when the training value accounts for 69% and the 27 elements with the best correlation and significance are involved, and the R² of training value and test value reaches 0.9992 and 0.7196 respectively.（2）Having the participation of 24 elements with the best correlation and significance, the optimal prediction model of PM₁₀ daily data in heavy pollution weather is achieved, and the R² of training value and test value are 0.9978 and 0.7792, respectively.（3）PM_2.5 optimal prediction model corresponds to the optimal five types of 700 hPa synthetic circulation situation.Its fourth and fifth category circulations and the first, third and fourth category circulations in the optimal six types of circulations accompanying PM₁₀ all belong to the high-pressure ridge control（impact） type and are prone to long-term continuous pollution, which is the main weather type affecting PM_2.5 and PM₁₀ in the Tianmu Mountains region.更多还原