Work

ExGAN-based assessment of extreme particulate matter risk and regional energy strategy for solar photovoltaic power generation stabilization

Year: 2026

Type: article

Abstract: The widespread adoption of solar photovoltaic (PV) systems is critical for sustainable energy transitions; however, extreme particulate matter (PM) significantly attenuates PV efficiency, especially in East Asia. To address this, we evaluate regional PM risks for PV generation to support optimal site selection and energy policy formulation. We propose a novel extreme generative adversarial network (ExGAN) framework coupled with a pixel-wise maximum ensemble strategy. While the conventional base ExGAN captures broad regional PM trends, it often misses localized spikes. By introducing a maximum-value distribution shift and ensembling the outputs, our refined model seamlessly integrates both broad geographical patterns and severe, localized extreme events. Consequently, the ensemble model achieved superior generative performance with a Fréchet inception distance of 3.96. The resulting risk map, based on an integrated average of 1-year to 100-year return levels, identifies metropolitan areas, such as Seoul and Gyeonggi, as high-risk zones, whereas southern coastal regions are classified as relatively low-risk environments. Cross-referencing these risks with existing PV capacities reveals that high-risk urban centers are highly exposed to extreme PM conditions, threatening the long-term stability of solar operations. Thus, for high-risk regions, a strategic shift toward complementary energy sources (e.g., offshore wind, waste-to-energy) alongside robust panel maintenance frameworks could be considered. Conversely, low-risk areas might benefit from maximizing PV efficiency through smart grids and expanding via floating solar deployments. Ultimately, by extending ExGAN to spatial environmental risk assessment and framing extreme PM as a critical operational constraint, we provide a data-driven framework for resilient regional renewable energy planning.

Source: Scientific Reports

DOI: 10.1038/s41598-026-65246-5

Authors: Taeyong Kwon, Kwangmin Lee, Sanghoo Yoon

Institution: Chonnam National University

Language: English

Topic: Solar Radiation and Photovoltaics

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

SDGs: Climate action, Affordable and clean energy, Sustainable cities and communities, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $2,690