Fast prediction of reservoir permeability based on EFS-LightGBM using direct logging data

Permeability estimation plays an important role in reservoir evaluation, hydrocarbon development, etc. Traditional methods have problems of time consuming and high cost. At present, the application of machine learning methods are more and more extensive, however, some machine learning models developed for permeability have fewer samples, requiring prior knowledge, and some parameters need to be calculated indirectly. To this end, based on a certain scale of permeability dataset, a hybrid method based on embedded feature selection and light gradient boosting machine (EFS-LightGBM) for reservoir permeability prediction is proposed. First, EFS is used to select features from the raw dataset, and then LightGBM is adopted to predict the permeability. The influence of feature selection threshold, base learners number and dataset size on prediction results is investigated. In addition, different feature selection and prediction models were compared, and the proposed method was also verified on other datasets. The experimental results show that the proposed method can effectively predict the permeability based on direct logging data. Its R2, RMSE and time are 0.9712, 0.5959 and 1.37s, respectively.