Prediction method and influence mechanism of effective thermal conductivity of complex fractured coal based on CT image reconstruction simulation and machine learning training

Abstract

The effective thermal conductivity (ETC) of coal body is a key parameter for geothermal design of abandoned mines, but due to the complex microscopic fracture structure of coal body after mining influence, the existing research lacks the prediction method of ETC the coal body considering the complex fracture structure. Therefore, in this study, finite element simulation was performed by inverse modelling of coal body CT slice data to form a dataset of 2200 sets. Six single machine learning (ML) prediction models of SVM, DT, KNN, BP, RF and XGBoost are trained by the dataset, and six stacking models are proposed on the basis of the single ML, and the prediction results are examined with the test set and the 3D modelling computation results by multiple indicators. The stacking prediction model with the best accuracy is finally screened, and SHapley Additive explanation (SHAP) analysis is performed to clarify the indicator influence mechanism. The resulting prediction procedure can batch calculate the ETC of coal bodies with real fissure structure in abandoned mines based on the coal mine core data, avoiding the sampling and testing of abandoned mines, and providing a reference for the engineering decision of heat extraction in the interior of abandoned mines.