Machine learning model combined with CEEMDAN algorithm for monthly precipitation prediction

Accurate forecasting of monthly precipitation is of great significance for national production, disaster prevention and mitigation, and water resources allocation management. However, it is difficult for individual models to accurately accomplish the task of predicting precipitation, and there is also the problem of insufficient accuracy of peak and trough prediction. Therefore, to solve this problem, this paper will provide a CEEMDAN-SVM-LSTM model that combines a fully adaptive noise ensemble empirical modal decomposition (CEEMDAN),a support vector machine (SVM) and the long short-term memory (LSTM) neural network. The CEEMDAN algorithm is first used to decompose the precipitation time series data into different modal components, the SVM model is then applied to the first modal component and the LSTM is applied to the remaining modal components. The precipitation data of Lanzhou city is taken as an example and brought into the model for testing and comparing with the performance of single LSTM model, differential integrated moving average autoregressive model (ARIMA), back propagation (BP) neural network model,support vector machine(SVM), extreme gradient boosting(XGBOOST), CEEMDAN-LSTM model and CEEMDAN-SVM model. After the experimental verification, the CEEMDAN-SVM-LSTM model effectively improves the fit between the observed and predicted values, overcomes the problem of low accuracy of peak and trough prediction, and significantly outperforms other models.