Research on precipitation prediction based on a complete ensemble empirical mode decomposition with adaptive noise–long short-term memory coupled model

Abstract

Scientific precipitation predicting is of great value and guidance to regional water resources development and utilization, agricultural production, and drought and flood control. Precipitation is a nonlinear, non-smooth time series with significant stochasticity and uncertainty. In this paper, a complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) with long short-term memory (LSTM) model is developed for predicting annual precipitation in Zhengzhou city, China, which is compared with a single LSTM model, an ensemble empirical mode decomposition–LSTM model, a complementary ensemble empirical mode decomposition–LSTM model, and a CEEMDAN–autoregressive integrated moving average and a CEEMDAN–recurrent neural network model. The results show that the mean absolute percentage error, root mean square error, and coefficient of determination of the coupled CEEMDAN–LSTM model are 2.69%, 17.37 mm, and 0.9863, respectively. The prediction accuracy is significantly higher than that of the other five models, indicating that the proposed model has high prediction accuracy and can be used for annual precipitation forecasting in Zhengzhou city. HIGHLIGHTS The CEEMDAN method adds adaptive Gaussian white noise in the decomposition, which effectively reduces the reconstruction error.; LSTM can effectively overcome the gradient explosion problem of recurrent neural networks and has significant advantages in handling long time series data.; The CEEMDAN–LSTM coupled model has good learning ability in dealing with nonlinear and non-smooth hydrological factor sequences.;

Keywords

• ceemdan
• lstm
• matlab
• precipitation prediction
• zhengzhou