Research on reverse identification algorithm of indoor pollution source based on machine learning

• Published in: E3S web of conferences Vol. 356; p. 4011
• Main Authors: Su, Xin, Long, Zhengwei, Xu, Yinyue
• Format: Journal Article Conference Proceeding
• Language: English
• Published: Les Ulis EDP Sciences 01-01-2022
• Subjects: Algorithms; Indoor air pollution; Indoor environments; Leakage; Learning algorithms; Machine learning; Occupational exposure; Occupational health; Optimization; Pollutants; Pollution sources; Working conditions
• ISSN: 2267-1242; 2555-0403; 2267-1242
• DOI: 10.1051/e3sconf/202235604011

Indoor sudden pollutant leakage brings environmental pollution and occupational exposure, so it is more and more important to obtain the location and identification of leakage sources. Through the forward method based on machine learning, this paper establishes a reverse traceability model for indoor multiple pollution sources. The POD method is used to obtain a large number of intermediate working condition data. The data pre-processing strategy of first normalization and then random forest feature screening can effectively improve the accuracy and generalization ability of the model. Based on a real environmental room case, model verification and sensor deployment optimization are carried out. The results show that the four sensors deployed in a specific location can achieve more than 95% positioning accuracy. In addition, the leakage possibility ranking component embedded in the model can effectively guide the staff to check the leakage points in turn, and the efficiency of three checks is as high as 99.91%.