基于经验模态分解的大气能见度预测

Atmospheric Visibility Prediction Based on Empirical Mode Decomposition

摘要: 经验模态分解(EMD)可以处理非线性、非平稳的时间序列数据，使之成为一组便于提取内部特征的分量。本文提出了一种结合经验模态分解(EMD)和AR模型的大气能见度预测方法。利用EMD将大气能见度数据分解为一组便于提取内部特征的本征模态函数(IMFs)。分别计算IMFs与原数据的相关系数，剔除相关性弱的本征模态函数以达到去噪目的。使用AR模型分别预测各个IMF，将各个预测值相加，得到最终的大气能见度预测值。为了验证所使用模型的预测性能，将只用AR模型的预测方法作为对比模型，并对比了两种模型的均方误差(RMSE)等指标，结果表明，结合了EMD的AR模型的预测性能优于只用AR模型的预测方法。

Abstract: Empirical Modal Decomposition (EMD) can deal with nonlinear, nonsmooth time series data into a set of components that facilitate the extraction of internal features. In this paper, we propose an atmospheric visibility prediction method that combines empirical modal decomposition (EMD) and AR modeling. The atmospheric visibility data are decomposed into a set of intrinsic modal functions (IMFs) that facilitate the extraction of internal features using EMD. The correlation coefficients between the IMFs and the original data were calculated separately, and the weakly correlated intrinsic modal functions were eliminated for denoising purposes. Each IMF was predicted separately using the AR model, and the predicted values were summed to obtain the final atmospheric visibility prediction. In order to verify the prediction performance of the model used, the prediction method using only the AR model was used as a comparison model, and the mean square error (RMSE) and other indexes of the two models were compared, and the results showed that the prediction performance of the AR model combined with the EMD was better than that of the prediction method using only the AR model.