基于SVM的沙尘暴预测模型

根据沙尘暴天气的特点和支持向量机（support vector machine，SVM）方法在解决小样本学习问题中的优势，提出基于SVM的沙尘暴预测模型．首先利用主成分分析法进行数据预处理，然后选择了径向基核函数，并通过分析惩罚参数和核参数对SVM分类器性能的影响，确定了参数的搜索空间，继而利用网格搜索法对其进行优化．在此基础上，构建并实现了基于SVM的沙尘暴预测模型．该模型与BP神经网络模型的运行结果对比表明，基于SVM的沙尘暴预报模型稳定性好，运行速度快，预报准确率提高了71．2％．

Sand-Dust Storm Forecasting Model Based on SVM

For the characteristics of the sand-dust storm weather and the advantages of support vector machine (SVM) in solving the learning problem with fewer samples,the sand-dust storm forecasting model based on SVM is proposed.The data is preprocessed by principal component analysis(PCA).Then the radial basic func- tion (RBF) kernel is chosen,and the search space of the penalty parameter and the kernel parameter is defined by analyzing the influence of the two parameters on the performance of SVM classifier.And the two parameters were optimized by grid search in the search space.Lastly,the sand-dust storm forecasting model based on SVM is constructed and implemented.Results comparison between the proposed model and the BP neural networks model show that the sand-dust storm forecost model based on SVM has a better stability,faster running speed and its forecasting precision ratio is increased by 71.2%.