A Deep Learning–Based Approach to Lightning Current Prediction Using Meteorological and Lightning Observations
기상·낙뢰 관측 데이터를 활용한 딥러닝 기반 낙뢰 전류 예측 모델 및 원자력 발전소 낙뢰 보호계통 설계 지원
Recent climate change has increased both the frequency and intensity of lightning events, posing growing risks to critical infrastructures such as nuclear power plants. Conventional lightning protection system (LPS) designs rely on fixed statistical assumptions that often fail to reflect regional climatic characteristics. This study proposes a data-driven lightning current intensity prediction framework for South Korea by integrating lightning observations from the KMA LINET system with meteorological data from ASOS stations. Machine learning models (Random Forest, XGBoost, and SVM) were compared with time-series deep learning models (GRU, Transformer, and TimesNet). Results demonstrate that time-series deep learning models outperform conventional approaches in prediction accuracy and robustness. The proposed framework provides adaptive, data-driven inputs that can support more reliable lightning protection system design.
한국어
최근 기후변화로 인해 낙뢰의 발생 빈도와 강도가 증가하면서 원자력발전소와 같은 국가 핵심 인프라의 안 전 위험이 확대되고 있다. 기존 낙뢰 보호 시스템은 국제 표준에서 제시하는 통계적 가정과 고정된 설계 전류 값을 기반으로 설계되어, 지역별 기후 및 지리적 특성을 충분히 반영하지 못하는 한계를 가진다. 본 연구는 이러한 한계 를 극복하기 위해 국내 원자력발전소 환경에 적합한 데이터 기반 낙뢰 전류 강도 예측 모델을 제안한다. 이를 위해 기상청(KMA)의 LINET 낙뢰 관측 자료와 ASOS 종관기상 관측 변수를 통합하여 분석하였다. Random Forest, XGBoost, SVM 등 전통적 머신러닝 기법과 GRU, Transformer, TimesNet 기반 시계열 딥러닝 모델을 적용하여 예측 성능을 비교 및 평가하였다. 실험 결과, 시계열 딥러닝 모델이 기존 머신러닝 기법 대비 더 높은 예측 정확도 와 안정성을 보였으며, 특히 낙뢰 전류의 시계열적 특성을 효과적으로 반영하는 것으로 나타났다. 본 연구에서 제안 한 프레임워크는 정적·경험적 설계 가정을 대체할 수 있는 정량적 근거를 제공함으로써, 원자력발전소 낙뢰 보호 설 계의 신뢰성과 현실 적용성을 향상시키는 데 기여할 수 있다.
목차
Abstract 요약 Ⅰ. Introduction Ⅱ. Background and Related Work 2.1 Conventional Lightning Protection System Design Standards 2.2 Limitations of Standard-Based Lightning Current Estimation 2.3 Lightning Observation and Statistical Analysis Studies 2.4 Machine Learning-Based Lightning Prediction Studies 2.5 Deep Learning and Time-Series Approaches 2.6 Research Gap and Motivation Ⅲ. Data and Methodology 3.1 Data Sources 3.2 Data Integration and Preprocessing 3.3 Feature Selection and Dataset Construction 3.4 Time-Series Data Construction 3.5 Dataset Partitioning and Normalization 3.6 Prediction Models 3.7 Evaluation Metrics Ⅳ. Experimental Results and Analysis 4.1 Experimental Environment 4.2 Statistical and Machine Learning-Based Prediction Results 4.3 Time-Series Deep Learning Prediction Results 4.4 Summary of Results Ⅴ. Conclusion 5.1 Summary of Findings 5.2 Academic Contributions 5.3 Practical Implications 5.4 Limitations and Future Work REFERENCES
키워드
낙뢰 전류 예측시계열 예측딥러닝 모델트랜스포머 기반 모델예측 성능 평가Deep learning models. Lightning current predictionPrediction performance evaluationTime-series forecastingTransformer-based models
저자
Yun-seob Lim [ 임윤섭 | Student, Department of Digital Convergence Engineering, Kumoh National Institute of Technology ]
Minjung Park [ 박민정 | Assistant Professor, Department of Business Administration, Kumoh National Institute of Technology ]
Corresponding Author
Ever since next generation convergence technology became one of the most important industries in the nation, computing professionals have encountered a growing number of challenges. Along with scholars and colleagues in related fields, they have gathered in avariety of forums and meetings over the last few decades to share their knowledge, experiences and the outcome of their research. These exchanges have led to the founding of the International Next-generation Convergence technology (INCA) on December 1, 2015. INCA was registered as an incorporated association under the Ministry of Information and Communications. The main purpose of the organization is to improve our society by achieving the highest capability possible in next generation convergence technology.
간행물
간행물명
차세대융합기술학회논문지 [The Journal of Next-generation Convergence Technology Association]