EEMD-Adaptive sparse Informer: Atmospheric temperature prediction based on the combination of EEMD and adaptive sparse Informer model

The temperature changes in the atmosphere are related to human activities. Therefore, precise prediction of atmospheric temperature is of great significance for human activities. Although the Informer model utilizes the Transformer structure to capture the mutual correlation and interaction between time steps in time series data, it ignores the inherent trend and periodicity of time series data. Based on the Informer capturing the correlation between time steps, we proposed decomposing the original sequence using the EEMD algorithm to obtain multiple IMF components. Then, we trained the improved Informer model with adaptive sparsity to get the overall trend of the time series. The EEMD-Adaptive sparse Informer model is proposed. The temperature data of Hangzhou city from the National Center for Environmental Information (NCEI) was verified. The results show that the EEMD-Adaptive sparse Informer model can improve the MSE of temperature prediction from 0.1292 to 0.1000, and the value of MAE can be increased from 0.2581 to 0.2404. This shows the EEMD-Adaptive sparse Informer prediction model effectively simulates and predicts atmospheric temperature.