The controlling factors and prediction model of pore structure in global shale sediments based on random forest machine learning

The rapid development of shale oil and gas promotes the plentiful research and characterization of micro-nano pore structure, and accumulates massive data, which is vital for fluid flow and storage. Although many interpretations based on various experimental methods and numerous shale specimens have been published, rare relatively complete compilation of shale pore structure has yet been made. Here, we report a new compilation of shale pore structure, comprising >350 marine, marine-terrestrial transitional and continental shale specimens of nearly all key formations and ages, of which collected samples were screened for effective analysis. This systematic review shows that shale sediments with different geological characteristics have obvious heterogeneity, multifarious micro-nano pores and complex pore size distribution (PSD). The shale PSD is divided into six types based on the morphological characteristics, and the compositions and thermal maturity of different PSD types of shale show obvious differences. Random forest (RF) machine learning (ML) further reveals that thermal maturity has the greatest effect on PSD classification, followed by organic matter (OM) abundance and mineral composition. Thermal maturation affects pore type and PSD. The high thermal maturation facilitates organic pores generation and micro-nano micropores connection. Additionally, the diagenesis related to thermal maturation will also affect the pore types, especially the organic acids formed during hydrocarbon generation, which will promote dissolution pores. OM and mineral affect the pore shape and type. OM abundance has positive and negative effects on the shale pore structure. The high OM abundance avails to the formation of more micro-nano micropores. However, the strong plasticity of OM will cause pore space collapse. Mineral composition mainly affects the pore structure via pore type. Quartz and feldspar mainly develop intergranular pores with a few fractures. Carbonate minerals are mainly dominated by dissolution pores. Interlayer pores are frequently formed in clay minerals. Based on the evaluation from RF ML, we suggest that the sedimentary environment and OM type have little influence on shale pore structure. A model based on RF ML is proposed to predict the PSD type of shale sediments from geological features. This model provides effective prediction for the six types of shale PSD. The limitation of valid data and the difference in data set size among different shale types constrain the accuracy of the prediction model. Supplementation of data and improvement of the RF model optimization algorithm in further studies will beneficially enhance the model accuracy. This study is a bold attempt, and provides a new time-saving and cost-saving method for the rapid prediction of shale pore structure based on geological analysis and ML. This work is instructive for understanding pore structure in shale sediments from basic geological information, and is valuable for further conserving resources and improving research efficiency.