Machine learning frameworks to analyze climate change impact on hydropower productivity

Abstract

Climate change profoundly impacts hydropower productivity, a cornerstone of renewable energy, necessitating advanced predictive tools for sustainable water-energy management. This study presents novel machine learning (ML) frameworks to forecast climate-induced variations in hydropower output by synergistically integrating climate, hydrological, and operational data with reanalysis datasets. Distinct from existing approaches, our methodology introduces unique contributions, including synthetic climate scenario generation via Generative Adversarial Networks (GANs), neural network-driven feature ranking to prioritize key climate variables, and robust preprocessing techniques such as outlier detection, normalization, and time-series feature engineering. Using a dataset of 650 records with 11 features from a hydropower plant in the Middle East, split into 70% training, 15% validation, and 15% testing subsets, we evaluated the performance of ARIMA, GAN, Autoregressive Deep Neural Network (ARDNN), and Long Short-Term Memory (LSTM) models using RMSE and R² metrics. The LSTM model outperformed the others, achieving an RMSE of 2892.61, a MAPE of 1.3237, and an R² of 0.9985, owing to its superior ability to capture long-term temporal dependencies. These advancements surpass traditional models by offering enhanced predictive accuracy and adaptability, enabling optimized resource management and bolstering the resilience of hydropower systems against climate variability, thus contributing significantly to global sustainable energy strategies.