Advanced machine learning schemes for prediction CO 2 flux based experimental approach in underground coal fire areas
Yongjun Wang,Mingze Guo,Hung Vo Thanh,Hemeng Zhang,Xiaoying Liu,Qian Zheng,Xiaoming Zhang,Mohammad Sh. Daoud,Laith Abualigah
DOI: https://doi.org/10.1016/j.jare.2024.10.034
IF: 12.822
2024-11-10
Journal of Advanced Research
Abstract:Highlights • This paper presents a highly effective approach aimed at reducing the necessity for costly and time-consuming experimental investigations of CO 2 flux in coal fire areas. • The study revealed that the NGRB scheme emerged as the most accurate method for estimating CO 2 flux in coal fire areas. • The study employed the estimation of SHAP values to assess the impact of input variables on the CO 2 flux estimation. • This study introduces an intelligent tool for predicting CO 2 flux in coal fires, which holds great potential for aiding in the evaluation of various disciplines within the mining sector. Introduction Underground coal fires pose significant environmental and health risks due to releasing CO 2 emissions. Predicting surface CO 2 flux accurately in underground coal fire areas is crucial for understanding the distribution of spontaneous combustion zones and developing effective mitigation strategies. In recent years, advanced machine learning techniques have shown promise in various carbon-related studies. This research uses an experimental approach to explore the power of advanced machine learning schemes for predicting CO 2 flux in underground coal fire areas. Objectives By leveraging the power of advanced machine learning schemes and experimental approaches, this research aims to provide valuable insights into CO 2 flux prediction in coal fire areas and inform environmental monitoring and management strategies. Methods The study involves the collection of an experimental dataset specific to underground coal fire areas, encompassing various parameters related to CO 2 flux and underground coal fire characteristics. Innovative feature engineering techniques are applied to capture the unique characteristics of underground coal fire areas and their impact on CO 2 flux. Different machine learning algorithms, including Natural gradient boosting regression (NGRB), Extreme gradient boosting (XGboost), Light gradient boosting (LGRB), and random forest (RF), are evaluated and compared for their predictive capabilities. The models are trained, optimized, and assessed using appropriate performance metrics. Results The NGRB model yields the best predictive performances with R 2 of 0.967 and MAE of 0.234. The novel contributions of this study include the development of accurate prediction models tailored to underground coal fire areas, shedding light on the underlying factors driving CO 2 flux. The findings have practical implications for delineating the spontaneous combustion zone and mitigating CO 2 emissions from underground coal fires, contributing to global efforts in combating climate change. Graphical abstract Download: Download high-res image (145KB) Download: Download full-size image
multidisciplinary sciences