A novel approach to precipitation prediction using a coupled CEEMDAN-GRU-Transformer model with permutation entropy algorithm

Abstract

Abstract The accurate forecasting of precipitation in the upper reaches of the Yellow River is imperative for enhancing water resources in both the local and broader Yellow River basin in the present and future. While many models exist for predicting precipitation by analyzing historical data, few consider the impact of different frequency sequences on model accuracy. In this study, we propose a coupled monthly precipitation prediction model that leverages the adaptive noise complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), gated recurrent unit neural network (GRU), and attention mechanism-based transformer model. The permutation entropy (PE) algorithm is employed to partition the data processed by CEEMDAN into different frequencies, with different models utilized to predict different frequencies. The predicted results are subsequently combined to obtain the monthly precipitation prediction value. The model is applied to precipitation prediction in four regions in the upper reaches of the Yellow River and compared with other models. Evaluation results demonstrate that the CEEMDAN-GRU-Transformer model outperforms other models in predicting precipitation for these regions, with a coefficient of determination R2 greater than 0.8. These findings suggest that the proposed model provides a novel and effective method for improving the accuracy of regional medium and long-term precipitation prediction.