Summary
With the accelerating global growth of renewable energy sources, electricity load forecasting systems are undergoing significant structural and methodological transformations. South Korea is experiencing a paradigm shift driven by the increasing penetration of non-metered solar photovoltaic (PV) generation, which introduces significant volatility and complexity. Although actual electricity consumption (hereinafter "Total Load") increases, the observable grid-based demand (Market/System Load) may appear to decline due to the self-consumption and injection from distributed PV systems, especially those that are not measured in real-time by the market
Read more Read less(Behind-The-Meter (BTM) and Korea Electric Power Corporation (KEPCO) Power Purchase
Agreements (PPA)).
To manage this discrepancy and ensure market transparency, Korea has adopted a dual-load reporting framework, separately publishing forecasts for Total Load and Market Load (System
Load). To maintain system reliability and optimize generation resource dispatch, Korea Power
System Operators (KPX) prioritized the development of the KPX Weekly Load Forecaster
(KWLF), a fully automated, web-based system that replaces previous offline, manual forecasting operations.
The methodology centers on a comprehensive approach: (1) enhancing renewable generation forecasting, and (2) refining the total system load forecasting model by accurately incorporating distributed generation effects. The forecasting platform emphasizes PV generation (due to
Korea's high capacity, currently 36 GW PV versus 2 GW wind). Crucially, the system estimates unmetered PV generation, particularly BTM PV, utilizing an artificial intelligence (AI)- based detection framework based on high-resolution imagery and the YOLO algorithm, significantly improving resource observability.
The final KWLF system projects the total system load first and then derives the market load by subtracting the estimated non-metered PV generation (BTM and PPA). It employs advanced machine learning techniques, including Long Short-Term Memory (LSTM) networks, eXtreme
Gradient Boosting (XGBoost), and the Relative Coefficient Method for special days (SDRS), combined through an Ensemble method for optimal accuracy. This comprehensive, data-driven methodology provides a robust framework for managing grid stability amidst increasing weather-dependent generation volatility.
Additional informations
| Publication type | Session Materials |
|---|---|
| Reference | C2_11419_2026 |
| Publication year | |
| Publisher | CIGRE |
| Country | Korea, Republic of (South Korea) |
| Study committees | |
| File size | 1 MB |
| Price for non member | 30 € |
| Price for member | 30 € |
Authors
SON Heung-Gu - Korea Power eXchange KOREA; KIM Hyunsu - Korea Power eXchange KOREA; LEE Kunsuk - Korea Power eXchange KOREA; LEE Chang Gun - Korea Power eXchange KOREA; SONG Kyung-bin - Soong-sil University KOREA
Keywords
