中国东部夏季雨型的人工神经网络集合预测

Ensemble prediction of summer rainfall patterns over eastern China based on artificial neural networks

摘要:

在BP(back propagation)人工神经网络方法的基础上,考虑到历史资料的个体差异及其年代际变化会影响到样本均值,由此使得中国东部夏季雨型模拟和预测效果产生差异,故引入交叉检验及集合预报思想,以改进人工神经网络独立预报方法。在利用不同历史样本资料建立人工神经网络模型,并进行交叉检验的同时,对预测年的雨型进行预测,可获得预测年的多次预测结果。该方法在中国东部夏季四类雨型的试验预报中表现出较好效果。

关键词:季节预测;集合人工神经网络方法;夏季雨型;中国东部

Abstract:

On basis of the back propagation artificial neural networks(BP ANN) method,and under consideration of the mean value differences caused by different sample collections for their individual differences and interdecadal variations,which may influence the simulation and prediction of summer rainfall patterns over eastern China,the cross validation and ensemble prediction ideas are adopted to improve the prediction method.An independent prediction is gotten during each cross validation test,and then many times cross validation test by the ANN model established based on different sample collections may give many times prediction results for the predicted year.Its performance seems good in the prediction test of four rainfall patterns over eastern China in summer.

Key words:seasonal prediciton;ensemble artificial neural networks method;summer rainfall pattern;eastern China