基于数据场的SVM技术在雷暴预报中的应用

针对天气预报中样本不平衡造成漏报率高的问题，提出一种基于数据场的C加权支持向量机（SVM）技术。该技术对不平衡天气数据进行分类，采用叠加数据场势值作为数据重采样依据，筛选出最利于SVM分类器学习的样本作为训练样本，结合C加权方法进行训练。实验结果证明，在样本数量较多且不平衡性显著的雷暴天气中，该技术能缩减训练集规模，减少漏报，提升预报系统的g-means值。

Application of SVM Technology Based on Data Field in Thunderstorm Report

Aiming at decreasing the rate of missing report caused by the imbalanced samples in weather report, this paper proposes a C weighted Support Vector Machine(SVM) technology based on data field. The technology classifies the imbalanced weather data, uses superimpose data filed potential value according as the data sample, the best samples for the SVM learning are filtered for training C weighted SVM. Experimental result proves that it can shrink the scale of training set, decrease the rate of missing report, boost the g-means in thunderstorm weather which has too many numbers of sample and prominent imbalanced property.