支持向量机在雷雨天气预报中的应用

Application of SVM in thunderstorm forecasting

支持向量机(SVM)是一种新型的机器学习方法。利用1999-2003年7月清远站每天08:00的探空资料,建立广州白云机场24 h内有无雷雨的SVM分类模型,进行相应的预报实验,实验结果显示对应的SVM分类模型效率高、准确率高,且泛化能力强,预报Ts评分非常理想,都达到80%以上;结果准确率并不会因为训练样本数目的减少而大幅度降低,具有良好的预报能力。对于某个特定的核函数,可通过调整误差惩罚参数C来得到性能最优的SVM。

Support vector machine （SVM） is a new machine learning technique. In this article, sounding data at UTC 00 in July during 1999 to 2003 from Qingyuan are used to establish the SVM classification model for thunderstorm prediction in 24 hours at Baiyun AirPort. The forecast experiments show that SVM classification model has satisfactory efficiency, accuracy and generalization performance with TS score over 80% ; Forecast accuracy doesnt decline with the reduction of training sample. Optimal SVM can be achieved by properly adjusting the penalty factor parameter.