ПОРІВНЯННЯ ПІДХОДІВ ДО ПРОГНОЗУВАННЯ НЕТТО-НАВАНТАЖЕННЯ ВІРТУАЛЬНИХ ОБ’ЄДНАНЬ ЕНЕРГЕТИЧНИХ ОБ’ЄКТІВ

• Date: 2025
• Main Authors: Мірошник, В.О., Сичова, В.В.
• Format: Article
• Language: Ukrainian
• Published: Інститут електродинаміки Національної академії наук України 2025
• Subjects: короткострокове прогнозування навантаження; агрегатори; локальні ринки електроенергії; нейронні мережі; віртуальні електростанції
• Online Access: https://prc.ied.org.ua/index.php/proceedings/article/view/412
• Journal Title: Proceedings of the Institute of Electrodynamics of the National Academy of Sciences of Ukraine

This study presents a comparative evaluation of three approaches to short-term net-load forecasting for virtual aggregations of heterogeneous energy facilities, comprising both building loads and photovoltaic units. The objective is to assess how different levels of data aggregation influence the forecasting accuracy of total consumption, PV generation, and the resulting net load. All experiments were conducted on the UCSD Microgrid dataset using an identical Transformer-based model, which allowed isolating the effect of data structure rather than model architecture. The results show that disaggregated bottom-up modelling yields the highest accuracy for building consumption, whereas partial aggregation—separate forecasting of total load and PV generation—provides the best performance for net-load prediction. PV generation remains the most challenging component due to structural asymmetry and extended zero-generation intervals; in this case, aggregated forecasts demonstrate lower energy error compared to bottom-up models. The findings provide practical guidance for selecting appropriate aggregation schemes and forecasting strategies in virtual power plants and emerging local energy markets. Ref. 14, fig. 4.