Weather-Aware Multi-Site Forecasting of Renewable Energy Generation and Electrical Load Using Deep Learning
Accurate renewable-generation and load forecasting depends not only on model choice, but also on weather variability, seasonal effects, site-specific operating conditions, and the amount and structure of available data. This project investigates how forecasting performance changes across sites, seasons, weather conditions, operating regimes, and dataset sizes.
The study compares plant-specific and pooled models as well as transfer-learning strategies. Persistence benchmarks, XGBoost, recurrent neural networks, and selected deep-learning architectures are evaluated with particular attention to cross-site generalization, robustness, limited historical data at newly deployed sites, and changing weather or operating regimes. The broader goal is to better understand the data requirements and transferability of forecasting models for renewable generation and electrical load.
Supervised by Prof. Dipti Srinivasan.