Coal permeability alteration prediction during CO₂ geological sequestration in coal seams: a novel hybrid artificial intelligence approach

Abstract: The technology of CO₂ geological storage while enhancing coalbed methane recovery has attracted significant attention. In this technology, the permeability alteration of coal after the injection of CO₂ is directly related to the injectability of CO₂, which is known as the CO₂ geological storage effect. Currently, laboratory tests are often carried out to obtain coal permeability, however, this method has several drawbacks due to the complicated testing procedure, high cost and complex sample preparation. To efficiently forecast the permeability alteration of coal in the process of CO₂ geological storage, this paper proposes an integrated new hybrid intelligent model of firefly algorithm (FA), sparrow search algorithm (SSA), and support vector machines (SVM). This FA-SSA-SVM hybrid intelligent model is trained and tested by a total of 154 data samples retrieved from the literature. The input variables include CO2 injection pressure, effective stress, coal rank, coal temperature and coal seam buried depth. The output variable is coal permeability. The evaluation indicators are R, MAE, RMSE, and MAPE. The results show that the FA-SSA-SVM prediction model have good potential for predicting the permeability alteration of coal during CO₂ geological storage. In addition, by comparing and analysing the evaluation indicators among the FA-SSA-SVM, SSA-SVM and L-MRA models, the model FA-SSA-SVM shows the highest accuracy, while the L-MRA model has the lowest accuracy. These research results can provide important guidance for promoting and applying CO₂ storage technology in coal seams. Article Highlights: The novel hybrid artificial intelligence model that integrates SVM, SSA, and FA is proposed to forecast coal permeability alteration.Firefly improved sparrow search algorithm can effectively optimize the hyper-parameters of SVM.The range of experimental/predictive permeability probability distribution of different prediction models are analyzed.