Spatiotemporal Modeling of Temperature Series Combining EMD Decomposition and Deep Learning
To address the complex spatiotemporal characteristics of atmospheric temperature data across different regions,a spatio-temporal modeling method based on Empirical Mode Decomposition(EMD)is proposed.A graph model is established using the lati-tude and longitude coordinates of the stations and the correlations of the station temperatures.The temperature series at each sta-tion undergo EMD decomposition,breaking the original data into several intrinsic mode functions(IMFs).By analyzing the correla-tion 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 re-sults,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.
万方数据
维普
