Genetic-Algorithm-Based Hierarchical Echo State Network for Temperature Prediction

Data-driven based temperature prediction methods are now highly competitive against numerical weather prediction (NWP) method. However, most deep learning-based prediction models focus on global weather forecasting, which requires substantial computational and data resources. Additionally, these models often exhibit lower accuracy in predicting temperatures in the upper atmosphere, which is crucial for studying the near space. To solve this problem, in this paper, the genetic-algorithm-based hierarchical echo state network (GA-HESN) is proposed. Firstly, the hierarchical architecture can fully extract information from the input sequence. Secondly, a parameter optimization strategy based on the genetic algorithm is proposed to further enhance the network's precision. Finally, experiments on the upper atmosphere temperature dataset for the Beijing region demonstrate the effectiveness of the model.