IntroductionUnderground coal fires pose significant environmental and health risks due to releasing CO2 emissions. Predicting surface CO2 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 CO2 flux in underground coal fire areas. ObjectivesBy leveraging the power of advanced machine learning schemes and experimental approaches, this research aims to provide valuable insights into CO2 flux prediction in coal fire areas and inform environmental monitoring and management strategies. MethodsThe study involves the collection of an experimental dataset specific to underground coal fire areas, encompassing various parameters related to CO2 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 CO2 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. ResultsThe NGRB model yields the best predictive performances with R2 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 CO2 flux. The findings have practical implications for delineating the spontaneous combustion zone and mitigating CO2 emissions from underground coal fires, contributing to global efforts in combating climate change.
Read full abstract