EMD分解与深度学习结合的温度序列时空建模  

Spatiotemporal Modeling of Temperature Series Combining EMD Decomposition and Deep Learning

作　　者：熊秋 彭龑 XIONG Qiu;PENG Yan(School of Automation and Information Engineering,Sichuan University of Science and Engineering,Yibin,Sichuan 644000,China;Key Laboratory of Artificial Intelligence,Yibin,Sichuan 644000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川宜宾644000 [2]人工智能四川省重点实验室,四川宜宾644000

出　　处：《宜宾学院学报》2024年第12期38-43,共6页Journal of Yibin University

基　　金：自贡市科技局科技计划资助项目(2018GYCX33)。

摘　　要：针对不同地区大气温度数据的复杂时空特征,提出一种基于经验模态分解(EMD)的温度序列时空建模方法.根据站点的经纬度坐标及站点温度的相关性建立图模型,对各个站点温度序列进行EMD分解,将原始数据分解为若干个模态函数;通过分析每个模态函数与原始数据的相关性,将不相关的模态函数分别相加,使用时空特征提取模块(GCN-LSTM)分别训练原数据和不相关数据,相减后输出,以捕捉数据的非线性时空关系.实验证明模型在多站点气温预测任务中,均方根误差较LSTM、GCN和GCN-LSTM模型分别下降了1.368、1.043、0.795,平均绝对误差分别下降了0.695、0.1625和0.1625.To address the complex spatiotemporal characteristics of atmospheric temperature data across different regions,a spatiotemporal modeling method based on Empirical Mode Decomposition(EMD)is proposed.A graph model is established using the latitude and longitude coordinates of the stations and the correlations of the station temperatures.The temperature series at each station undergo EMD decomposition,breaking the original data into several intrinsic mode functions(IMFs).By analyzing the correlation between each IMF and the original data,uncorrelated IMFs are summed separately.A spatiotemporal feature extraction module(GCN-LSTM)is then used to train the original data and the uncorrelated data separately.The output,obtained by subtracting the results,captures the nonlinear spatiotemporal relationships in the data.Experiments demonstrate that the model achieves a root mean square error reduction of 1.368,1.043,and 0.795 compared to the LSTM,GCN,and GCN-LSTM models,respectively,and a mean absolute error reduction of 0.695,0.1625,and 0.1625,respectively,in multi-station temperature prediction tasks.

关 键 词：经验模态分解(EMD) 图卷积网络(GCN) 长短期记忆网络(LSTM) 温度序列时空建模 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]