Data-driven Point Prediction of Water Inflow for Cascade Small Hydropower Reservoirs Considering Multi-dimensional Impacts

This paper focuses on the power generation capacity prediction of small hydropower and the inflow prediction for cascade small hydropower reservoirs. Small hydropower exhibits unique characteristics that differentiate it from large hydropower in terms of regulation, capacity, location, and distribution, necessitating specific prediction methods. Reservoir inflow prediction is crucial for water resource allocation and hydropower station operation optimization. Traditional methods rely on hydrological models, which are complex and data-intensive. Alternatively, data-driven and machine-learning approaches have gained popularity due to advancements in computer performance. This paper proposes a data-driven point prediction solution for water inflow, incorporating multi-dimensional data such as historical and forecast weather information. The solution utilizes deep learning models, particularly recurrent neural networks and their variants like Variational Mode Decomposition Long Short-Term Memory Neural Networks (VMD-LSTM), for daily and monthly inflow predictions. Evaluation results demonstrate the effectiveness of this approach in enhancing prediction accuracy.