Machine learning based quantitative consequence prediction models for toxic dispersion casualty

• Published in: Journal of loss prevention in the process industries Vol. 81; p. 104952
• Main Authors: Jiao, Zeren, Zhang, Zhuoran, Jung, Seungho, Wang, Qingsheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-02-2023
• Subjects: Consequence modeling; Gradient boosting; Machine learning; Toxic dispersion; Consequence modeling; Toxic dispersion; Machine learning; Gradient boosting
• ISSN: 0950-4230
• DOI: 10.1016/j.jlp.2022.104952

Incidental release of toxic chemicals can pose extreme danger to life in the vicinity. Therefore, it is crucial for emergency responders, plant operators, and safety professionals to have a fast and accurate prediction to evaluate possible toxic dispersion life-threatening consequences. In this work, a toxic chemical dispersion casualty database that contains 450 leak scenarios of 18 toxic chemicals is constructed to develop a machine learning based quantitative property-consequence relationship (QPCR) model to estimate the affected area caused by toxic chemical release within a certain death rate. The results show that the developed QPCR model can predict the toxic dispersion casualty range with root mean square error of maximum distance, minimum distance, and maximum width less than 0.2, 0.4, and 0.3, which indicates that the constructed model has satisfying accuracy in predicting toxic dispersion ranges under different lethal consequences. The model can be further expanded to accommodate more toxic chemicals and leaking scenarios.
•A toxic chemical dispersion casualty database was constructed.•A quantitative property-consequence relationship (QPCR) model was build.•The QPCR model can estimate the affected area caused by toxic chemical release.•The QPCR model can be extended to more toxic chemicals and leaking scenarios.