Author(s): Su Han Nam; Si Yoon Kwon; Jae Yeon Lim; Young Do Kim

Linked Author(s): Young Do Kim

Keywords: Chemical accident; Chemical spill; Alternative indicator; Machine learning

Abstract: Chemical accidents in water can occur due to natural and man-made causes, and when such chemical accidents occur, it can cause changes in the aquatic environment and adversely affect the ecosystem or humans, so a prompt initial response is required. Various types of chemical spill accidents occur frequently around the world. When a chemical accident occurs, laboratory-based analysis is carried out, but this analysis has limitations in initial response. For the initial response, an index that is easy to measure in the field should be used, and information on the leaked chemical should be obtained using this index. In addition, if machine learning techniques are applied to the measured data, it will be helpful in the initial response in the event of a chemical accident. In this study, 26 chemical substances were used, and pH and EC according to concentration changes were established as a database. The database was represented by the pH-EC relation curve, and chemicals were classified into groups based on the spatial distribution and trend of the data. Based on the established database, the best performing machine learning method is selected by applying the machine learning methods such as Decision Tree, Random Forest, Gradient Boosting, and XG Boosting to evaluate the performance of each algorithm. If the machine learning technique selected based on the results of this study is used, information on the harmful chemicals leaked in the case of a chemical accident in the perfume water environment can be provided, and it is judged that it can be used as basic data for a prompt response.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221168

Year: 2022