基于改进EMD算法和BP神经网络的SST预测研究

发布时间：2018-04-29 19:13

本文选题：集合经验模态分解 + 互补集合经验模态分解　；参考：《气候与环境研究》2017年05期

【摘要】：海洋表面温度(Sea Surface Temperature,SST)具有非平稳、非线性的特征,直接将处理平稳数据序列的方法应用到非平稳非线性特征明显的序列上显然是不合适的,预测的误差将会很大。为了提高预测精度,更好地解决非平稳非线性序列预测的问题,本文以东北部太平洋(40°N~50°N、150°W~135°W)区域的月平均海洋表面距平温度为例,首先分别应用集合经验模态分解(EEMD)和互补集合经验模态分解(CEEMD)方法将SST分解为不同尺度的一系列模态分量(IMF),再运用BP(Back Propagation)神经网络模型对每一个模态分量进行分析预测,最后将各IMF预测结果进行重构得到SST的预测值。数值试验的结果表明,CEEMD分解精度比EEMD分解精度高,CEEMD提高了基于BP神经网络的预测精度。系列试验统计分析说明应用这种方法对SST的1年预测是有效的。
[Abstract]:The sea surface temperature (Surface) has the characteristics of nonstationary and nonlinear. It is obviously inappropriate to apply the method of dealing with stationary data series to the sequence with obvious nonstationary nonlinear characteristics, and the prediction error will be very large. In order to improve the prediction accuracy and solve the problem of nonstationary nonlinear series prediction, this paper takes the monthly mean ocean surface anomaly temperature in the northeast Pacific Ocean as an example. Firstly, the SST is decomposed into a series of modal components of different scales by using the methods of set empirical mode decomposition (EEMD) and complementary set empirical mode decomposition (CEEMD), respectively. Then the BP(Back propagation neural network model is used to analyze and predict each modal component. Finally, the IMF prediction results are reconstructed to get the SST prediction value. The results of numerical experiments show that the accuracy of CEEMD decomposition is higher than that of EEMD decomposition, and the prediction accuracy based on BP neural network is improved. The statistical analysis of series experiments shows that this method is effective for 1 year prediction of SST.
【作者单位】：中国人民解放军理工大学理学院;
【基金】：国家自然科学基金项目91530204、41430426~~
【分类号】：P731.31

【相似文献】