Prediction of the shale gas permeability: A data mining approach

Abstract

One of the most important parameters affecting shale gas extraction is the gas permeability of shale. Because there are many influencing factors and the mechanism of interaction is complex, it is difficult to accurately predict the gas permeability of shale. In this paper, a new machine learning model is proposed by combining Mind Evolutionary Algorithm (MEA) and Adaptive Boosting Algorithm-Back Propagation Artificial Neural Network (ADA-BPANN), which predicted the gas permeability of cement mortar with different moisture contents under different stress conditions based on the results of 616 laboratory gas permeability experiments. This is the first time that a combination of MEA and ADA-BPANN algorithms has been used to predict shale gas permeability. Compared to the traditional machine learning algorithms such as Particle Swarm Optimization Algorithm (PSO) and Genetic Algorithm (GA) optimized ADA-BPANN. The excellent performance of MEA optimized ADA-BPANN has been verified. This novel algorithm has higher prediction accuracy, shorter training time, and can avoid problems such as local optimization and overfitting. Secondly, the sensitivity of the parameters is analysed by using the novel model, and the results show that the parameter with the greatest influence on gas permeability is relative moisture content, followed by confining pressure, seepage pressure and confining pressure loading/unloading stage. The present research shows that the MEA optimized ADA-BPANN model has great potential for estimating the stress-dependent gas permeability of shale with different moisture contents. It is very helpful for the shale gas exploitation.