支持向量机与卡尔曼滤波集合的西太平洋副热带高压数值预报误差修正

基于T106数值预报产品资料,提出了支持向量机和卡尔曼滤波相结合的方法来进行夏季西太平洋副热带高压数值预报的误差修正与预报优化。首先采用支持向量机方法建立了西太平洋副热带高压面积指数的误差修正模型。基于支持向量机预报优化模型尽管有比较好的拟合精度和预报效果,但与实际副热带高压指数尚有一定的差异。究其原因,除预报对象(副热带高压)本身比较复杂、模型优化因子不够充分以及数值预报误差自身的随机性以外,优化模型的输入、输出基本上是一个静态映射结构,因此前一时刻的预测误差难以得到有效的反馈、调整和修正。为考虑前一时刻预报误差的反馈信息,动态跟踪副高的变化趋势,随后引入卡尔曼滤波方法建立支持向量机-卡尔曼滤波模型,对支持向量机模型的输出结果作进一步的调整和优化。试验结果表明,该方法模型的预报优化效果优于T106数值预报产品以及单纯的神经网络修正模型和卡尔曼滤波修正模型的优化效果,能够较为客观、有效地修正西太平洋副热带高压指数的数值预报误差,改进和优化西太平洋副热带高压的数值预报效果。该方法为副热带高压等复杂天气系统和要素场预报提供了一种新的思路,表现出较好的应用前景。

ERROR-CORRECTING OF THE AREA INDEX OF SUBTROPICAL HIGH IN THE T106 NUMERICAL PREDICTION BASED ON SUPPORT VECTOR MACHINE-KALMAN FILTER MODEL

Based on the T106 NWP product information,the T106 numerical forecast error of western Pacific subtropical high was corrected and optimized using the methods of support vector machine(SVM) and Kalman filter.An error-correcting model for the area index of the western Pacific subtropical high was first established with the SVM method.Despite of its good fitting accuracy and forecast results,there were still many differences between forecast results of the SVM model and actual results.There were many reasons for the differences.In addition to complex forecast object itself,inadequate model optimization factors and random numerical forecast errors,the SVM forecast model has basically a static mapping structure,therefore the anterior forecast errors were difficult to be effectively feedbacked.In consideration of the anterior forecast errors,the Kalman filter was introduced to establish the Kalman-support vector machine model to further optimize and adjust the output of the support vector model.The testing results show that the Kalman-support vector model can objectively and effectively correct the T106 numerical forecast error of western pacific subtropical high,and is superior to the T106,BP,and Kalman models in high forecast accuracy,fast training,high generalization capability and easy modeling,thus providing a new method for the forecast of the complex weather system such as subtropical high etc.