# SolarFlux Predictor: A Novel Deep Learning Approach for Photovoltaic Power Forecasting in South Korea

## Bibliographic information / 논문 정보

- Authors as listed: Hyunsik Min, Seokjun Hong, Jeonghoon Song, Byeoungmin Son, Byeongjoon Noh\*, Jihoon Moon\*
- Archive year: 2024
- Category: International Journal
- Status: Published
- Venue / registration record: Electronics 13(11), 2071, 2024.
- Public lab record: https://vita.hanyang.ac.kr/publications/2024-4auhmd-solarflux-predictor-a-novel-deep-learning-approach-for-photovoltaic-power-forecasting-in-south-k/index.html
- Text record: https://vita.hanyang.ac.kr/publication-detail/record-043177d0cf378937.txt
- Markdown record: https://vita.hanyang.ac.kr/publication-detail/record-043177d0cf378937.md
- Original source: https://doi.org/10.3390/electronics13112071
- DOI: 10.3390/electronics13112071

## Research summary / 연구 요약

SolarFlux Predictor는 태양광 발전량의 시간적 변동을 예측하기 위해 TCN과 다중 헤드 self-attention을 결합한 모델입니다. 발전량·기상·시간 정보를 전처리하고, teacher forcing과 Optuna 기반 설정 탐색을 적용해 여러 시점의 발전량을 예측합니다. 국내 9개 지역의 자료에서 2017–2018년을 학습에 사용하고 2019년을 예측해 TCN 계열 비교 모델보다 낮은 예측 오차를 보고했습니다. 검증은 해당 지역과 기간의 자료에 근거하며, 복잡한 모델의 계산 비용과 다른 기후·지역으로의 확장 가능성은 추가 검토 과제입니다.

SolarFlux Predictor forecasts photovoltaic power by combining temporal convolutional networks with multi-head self-attention. Its workflow integrates power-generation, weather, and temporal features, data preprocessing, teacher forcing during training, and Optuna-based hyperparameter optimization. Evaluation covers nine South Korean regions, using 2017–2018 for training and 2019 for prediction. Comparisons with standard and attention-enhanced TCN variants report lower forecasting errors for SolarFlux across the regional datasets. The study supports multistep prediction under these historical conditions. The paper also discusses computational demands and the need to assess adaptation to other regions and climates before extending its conclusions.

Topics: photovoltaic forecasting; temporal convolutional network; self-attention; teacher forcing; SolarFlux

## Research areas / 연구 분야

AI · 인공지능 / Artificial intelligence; 딥러닝 / Deep learning; 에너지 / Energy

## Source and scope / 출처와 범위

This is a lab-maintained overview, not an abstract supplied by the publisher. The archive year and author spelling follow the publication list; exact dates and affiliations are not inferred.

출판된 논문 전문의 방법론, 실험 결과와 결론을 확인하여 작성한 연구 요약입니다. 결과는 논문에서 평가한 자료와 조건의 범위로 해석합니다.

This research summary is based on the methods, experimental results, and conclusions of the published full text. Findings are interpreted within the data and conditions evaluated in the paper.
- [출판사 논문 및 서지정보 (DOI)](https://doi.org/10.3390/electronics13112071)

## Related research / 관련 연구

- [SolarNexus: A deep learning framework for adaptive photovoltaic power generation forecasting and scalable management](https://vita.hanyang.ac.kr/publication-detail/2025-solarnexus-a-deep-learning-framework-for-adaptive-photovoltaic-power-generation-forecastin.html)
- [Stacking deep transfer learning for short-term cross building energy prediction with different seasonality and occupant schedule](https://vita.hanyang.ac.kr/publication-detail/2022-stacking-deep-transfer-learning-for-short-term-cross-building-energy-prediction-with-diffe.html)
- [In-depth analysis of energy efficiency related factors in commercial buildings using data cube and association rule mining](https://vita.hanyang.ac.kr/publication-detail/2017-in-depth-analysis-of-energy-efficiency-related-factors-in-commercial-buildings-using-data-.html)