| Title |
GATConv-LSTM Insolation Prediction Model with Spatio-temporal Interactions |
| Authors |
장영준(Young-Jun Jang) ; 이석호(Seok-ho Lee) ; 김지홍(Kim Jeehong) ; 정길도(Kil To Chong) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.3.38 |
| Keywords |
Artificial intelligence; Renewable energy; Time Series Data; Spatiotemporal analysis approach; Graph neural networks |
| Abstract |
Solar insolation, which is critical for solar power generation, is complex and non-linear, driven by a variety of external environmental factors, including terrain, humidity, temperature, precipitation, wind speed, and wind direction. Existing time-based forecasting models focus only on temporal patterns and suffer from overfitting in certain regions, resulting in poor performance in other regions. In this paper, we propose a spatio-temporal based insolation forecasting model that combines Graph Attention Convolution (GATConv) and Long Short-Term Memory (LSTM) to address these issues. GATConv learns the spatial interactions between weather stations, and LSTM predicts insolation changes by reflecting temporal patterns. By considering both temporal and spatial characteristics, the model maintains the stability of prediction performance under various weather conditions and topographical variations, and demonstrates improved generalization performance. |