논문 상세보기
pp.354-360
This study proposes a novel diagnostic methodology combining mobile measurement using selected ion flow tube mass spectrometry (SIFT-MS) and explainable artificial intelligence (XAI) to effectively monitor and diagnose localized highozone (O3) events in industrial complexes. The methodology was applied to a highconcentration ozone episode (maximum 94.0 ppb) observed in the Hwaseong Bio Valley, an industrial complex. A nonlinear regression model based on the Random Forest algorithm was developed to quantify the contribution of precursor species. Specifically, to precisely diagnose the individual contributions of volatile organic compounds (VOCs), which are critical determinants of ozone formation, a modeling approach centered on VOCs was employed by excluding inorganic precursors (NOx). Contrary to traditional ozone formation potential (OFP) analysis, which prioritized high-reactivity alkenes such as propene, the AI model identified cyclohexane and butanone (MEK) as the key drivers positively correlated with ozone concentration fluctuations. This discrepancy is attributed to the “abundance effect,” where atmospheric partial pressures of organic solvents, extensively emitted from pharmaceutical and bio-industrial processes, overwhelm the differences in chemical reactivity of individual species. The findings suggest that AI techniques can interpret the nonlinearity of complex photochemical reactions based on observational data, serving as a complementary site-specific diagnostic tool to existing property-based assessments (e.g., MIR). Consequently, future air quality policies should shift from uniform regulations to a more targeted approach, utilizing the proposed methodology to establish precise emission tracking and management systems.
1. 서 론
2. 연구 방법
2.1 연구 대상 지역 및 측정 개요
2.2 이동측정시스템 구성 및 운용 방법
2.3 데이터 전처리 및 통합 절차
2.4 인공지능(AI) 기반 분석 방법
3. 결과 및 고찰
3.1 오존 및 주요 VOCs의 시계열 특성
3.2 AI기반 VOCs 기여도 및 SHAP 분석 결과
3.3 OFP 분석과 AI 결과의 상관 비교
4. 결 론
References
