目前雾霾污染日益严重, 威胁到了环境保护和人类健康, 需要对雾霾 天气进行预测. 通过对多个支持向量机(SVM)进行选择性集成, 克服单个SVM不稳 定的缺点, 提出了基于相异度的SVM选择性集成雾霾天气预测方法(DSE-SVM). 首先采用高斯核SVM独立训练出多个个体SVM; 其次计算出个体SVM的相异度, 剔除相异度最大的个体SVM; 最后运用多数投票算法对剩余的SVM进行集成, 并进行了理论分析. 通过对北京、上海和广州三地区近两年的雾霾数据进行实验分析, 实验结果表明DSE-SVM方法预测性能更优, 具有较高的稳定性和可信性.

Haze is becoming increasingly serious, which is harmful to environmental protection and human health, and it is necessary to predict in advance. Selective ensemble of SVM based on disagreement measure for haze forecast is proposed by selective ensemble of SVMs, which also overcomes instability of single SVM. Firstly, train some based classifiers by SVM based Radial Basis Function independently; Secondly, calculate disagreement measure of every SVM and the SVM, which has maximum disagreement measure, needs to be removed. Finally, the rest of SVMs are grouped by majority voting, and theoretical basis of DSE-SVM is analyzed. By analyzing experiments on haze data set of Beijing, Shanghai and Guangzhou for nearly two years, experimental results show that the performance of DSE-SVM is superior to the methods, has relatively high stability and credibility.