Prediction of water levels in large reservoirs base on optimization of deep learning algorithms

Water level forecasting is a critical technique for reservoir water resource management and flood early warning. This study addresses the limitations of the traditional Long Short-Term Memory (LSTM) network in terms of accuracy and generalization when handling complex hydrological data. To improve the precision and stability of LSTM in water level forecasting, four optimization algorithms—African Vulture Optimization Algorithm (AVOA), Cuckoo Search (CS), Firefly Algorithm (FA), and Grey Wolf Optimizer (GWO)—were introduced to optimize the LSTM model. The study employed a self-developed RIL scoring standard to comprehensively evaluate the models’ performance. The results show that all optimized models significantly outperformed the traditional LSTM model. Among them, the GWO-LSTM achieved the best performance in terms of accuracy, with a Mean Absolute Error (MAE) of 0.1043 m, a Root Mean Square Error (RMSE) of 0.1402 m, and the highest RIL score of 2.4364. The study confirms the effectiveness of combining optimization algorithms with LSTM models in water level forecasting, offering a method to significantly improve prediction accuracy. It also provides new directions for enhancing the model’s generalization capability and adaptability. Accurate water level forecasting in large reservoirs not only provides a scientific basis for reservoir management but also has significant theoretical and practical implications for flood control, disaster mitigation, ecological protection, and the sustainable use of water resources.