Research on flood season runoff forecast based on a novel combination machine learning method

Hydrological forecasting is of great significance for flood control and disaster reduction, water resource management, hydropower energy production, and environmental protection. This article proposes a novel approach that combines an artificial intelligence algorithm and a decomposition algorithm. The proposed method first uses a Seasonal Decomposition (SD) algorithm to decompose the runoff series into subcomponents including trend, seasonal, and residual components. Then, the Simple Recurrent Unit (SRU) algorithm is used to make predictions for each subcomponent, which are then integrated into the final prediction result. The proposed model was tested on the Dongjiang dataset along with several other common machine learning prediction models including Gated Unit Recurrent (GUR), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Backpropagation (BP), and Linear Regression (LR). The results showed that the proposed model outperformed the other models, with a coefficient of determination R2 of 88.94 and a Nash-Sutcliffe efficiency NSE of 0.936. The proposed model is particularly effective at predicting flood season runoff sequences.