Publicação
Forecasting Hybrid Power Plant Generation: The case of Joint and Isolated Solar and Wind Forecasting in Hybrid Power Plants in the Iberian Peninsula
| Resumo: | The rapid expansion of wind and solar generation in the Iberian Peninsula has reshaped power systems operations, elevating short-term forecasting to a core risk management tool. In hybrid solar–wind plants connected at a single interconnection point, uncertainty arises not only from individual resource variability, but from technological interaction under shared grid constraints. Yet solar and wind generation are typically forecasted independently and aggregated only a posteriori. This dissertation examines whether jointly forecasting reduces predictive uncertainty relative to isolated approaches. Using real operational data from hybrid plants operated by EDP in Portugal and Spain, the study adopts a system-level perspective centered on net export at the interconnection point. A comparative framework evaluates isolated and joint formulations under identical data, training, and evaluation conditions. Statistical baselines and structured deep learning architectures are tested with emphasis on NHITS due to its multi-scale temporal representation. Results show that joint forecasting reduces Mean Absolute Error (MAE) by approximately 20.54% compared to isolated formulations. This improvement does not increase the intrinsic predictability of solar or wind generation but emerges from partial error compensation across technologies with distinct temporal dynamics. By mitigation extreme deviations, the joint formulation reduces exposure to tail events that drive operational and market risk in renewable dominated systems. These findings demonstrate that forecasting performance in hybrid plants depends not only on model sophistication but on how the forecasting problem is structured and evaluated, positioning hybrid forecasting as a system-level uncertainty management challenge under realistic industrial constraints. |
|---|---|
| Autores principais: | Almeida, Leonor Mergulhão Alves Baptista de |
| Assunto: | Hybrid Renewable Power Plants Solar-Wind Complementarity Joint Forecasting Net Export Forecasting NHITS Renewable Energy Uncertainty ReCycle |
| Ano: | 2026 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | The rapid expansion of wind and solar generation in the Iberian Peninsula has reshaped power systems operations, elevating short-term forecasting to a core risk management tool. In hybrid solar–wind plants connected at a single interconnection point, uncertainty arises not only from individual resource variability, but from technological interaction under shared grid constraints. Yet solar and wind generation are typically forecasted independently and aggregated only a posteriori. This dissertation examines whether jointly forecasting reduces predictive uncertainty relative to isolated approaches. Using real operational data from hybrid plants operated by EDP in Portugal and Spain, the study adopts a system-level perspective centered on net export at the interconnection point. A comparative framework evaluates isolated and joint formulations under identical data, training, and evaluation conditions. Statistical baselines and structured deep learning architectures are tested with emphasis on NHITS due to its multi-scale temporal representation. Results show that joint forecasting reduces Mean Absolute Error (MAE) by approximately 20.54% compared to isolated formulations. This improvement does not increase the intrinsic predictability of solar or wind generation but emerges from partial error compensation across technologies with distinct temporal dynamics. By mitigation extreme deviations, the joint formulation reduces exposure to tail events that drive operational and market risk in renewable dominated systems. These findings demonstrate that forecasting performance in hybrid plants depends not only on model sophistication but on how the forecasting problem is structured and evaluated, positioning hybrid forecasting as a system-level uncertainty management challenge under realistic industrial constraints. |
|---|