Abstract
Flow is one of the important hydrological elements to study the water ecology and water environment of rivers in nature. Predicting flow is crucial for gathering valuable research data to aid in flood prevention, mitigation efforts, and the sustainable harnessing and utilization of water resources in the basin. To enhance the accuracy of flow prediction, a novel approach has been proposed. This methodology integrates the Adaptive Noise Complete Ensemble Empirical Modal Decomposition (CEEMDAN) with a Long and Short-Term Memory (LSTM) model, further refined through the application of the Sparrow Search Algorithm (SSA). The result is a powerful and innovative Combined Runoff Prediction Model, referred to as CEEMDAN-SSA-BiLSTM. This integrated model aims to provide more reliable predictions for both long and short-term runoff scenarios, contributing to more effective water resource management and environmental preservation in the basin. The daily flow trends from 2016 to 2022 were analyzed at four hydrological stations, namely Huayuankou, Jiahetan, Gaocun, and Lijin. The overall process is to use 80% daily flow data trained to predict 20% daily flow. Combined with the evaluation indexes used, the final series of results obtained are compared with the prediction results of several models, such as LSTM, BiLSTM, and CEEMDAN-BiLSTM, in multiple ways. The ultimate comparative outcomes demonstrate that the CEEMDAN-SSA-BiLSTM coupling exhibits a notable level of accuracy in forecasting daily flow. It has less error compared to several other models.