(95a) A Novel Approach for VOC Emission Characteristics Identification Based on Deep Learning and Its Application in Source Tracing of a Chemical Industrial Park

Refined VOC emission characteristics are an important prerequisite for accurate source tracing in a chemical industrial park. The navigation monitoring data contains a lot of pollution characteristic information that is not intuitively displayed, which needs to be further excavated. This study proposed a novel approach for VOC emission characteristics identification based on the CAM of CNN, which was applied in a typical fine chemical industrial park with the navigation monitoring data. With the method, 5 key factories’ characteristic VOCs were identified. Then, the VOC source tracing work of the park for 4 time periods in a day was carried out by PMF. 7 factor profiles were calculated and identified as pharmaceutical factories represented, pesticide factories represented, dye factories with large use of solvents, power plants, et al. The results illustrated that this approach can fully exploit the emission characteristics and be applied to existing source tracing models to improve the degree of refinement.