Long-term continuous monitoring of volatile organic compounds (VOCs) is essential for atmospheric pollution control, environmental health assessment, and climate change mitigation. Gas chromatography (GC) serves as the primary analytical technique for precise VOCs quantification in prolonged atmospheric monitoring programs. However, accurate chromatographic peak segmentation critically depends on labor-intensive manual analysis by experienced specialists, particularly when dealing with overlapping peaks caused by instrument degradation or prolonged monitoring. This study presents an innovative artificial intelligence framework combining a peak identification model (ResGRU) and time-series semantic segmentation model (UGRU) for automated gas chromatographic peak segmentation. The UGRU model innovatively improves traditional semantic segmentation frameworks by replacing convolutional feature extractors with sequential algorithms based on the temporal dynamics characteristics of chromatographic signals, achieving superior performance across four monitoring sites in Shanghai, Hubei, and Jiangsu, China, with over 85% of samples achieving absolute percentage errors below 10% and over 90% achieving errors below 20%. Cross-site transfer experiments demonstrated robust generalizability, requiring only 512 target-site samples to achieve effective adaptation through fine-tuning strategies. This work provides a transformative solution for large-scale automated VOC monitoring networks, significantly reducing manual analysis costs while enhancing data standardization, thereby enabling more effective atmospheric pollution management.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号