Hydropower Plant Generation Forecasting using Long Short-Term Memory (LSTM) for Optimizing Water Utilization

Abstract

Hydropower plants face significant challenges with water availability as climate change increases weather uncertainty, affecting water usage efficiency. To improve the efficiency of water usage, decision-making should be based on long-term data and prediction methods beyond just water levels and weather forecasts. Therefore, improvements are needed in decision-making regarding operating patterns to increase hydropower efficiency. Long Short-Term Memory (LSTM) methods in water use prediction offer an innovative approach to increase efficiency and reduce waste of water resources. LSTM, a variant of Recurrent Neural Network (RNN), can recognize long-term patterns and dependencies in time series data, making it ideal for predicting fluctuating energy production capacity. By applying LSTM to predict production energy, a more accurate and reliable prediction model could be obtained. The model is designed to enhance water use predictions, optimize hydropower operations for efficient resource management, and support scientific basis decision-making based on data in water management. Experimental results show a lower error according to its predictive capacity with the normalized RMSE of 0.06170 and 0.96391 R2 value. The results can then be used in real operation scenarios. It is concluded that the LSTM model is a good strategy for the forecast of water flow for the study of hydroelectric turbine efficiency. This paper discusses the forecasting strategy of production capacity at PM Noor, which uses the LSTM method to assist operation scenarios.