关键词: 储层参数预测, 迁移学习, SMOTE, 随机森林, 神经网络

Abstract: To address the issue of traditional methods neglecting flow mechanisms and parameter correlations,a prediction model of reservoir parametert hat integrates seepage theory with transfer learning is proposed.By using the oversampling algorithm of SMOTE,the issue of imbalanced samples is effectively addressed.The discriminative model of lithology andseepage capacity is established by using random forest,which provides the information of seepage mechanism for reservoir parameter predicting.Combined with parameter correlation,the technology of transfer learningis used to build a reservoir parameter prediction model.The results show that the correlation analysis between reservoir parameters can be conducted by introducing lithology and seepage capacity discrimination technology,which can effectively improve the prediction accuracy of reservoir parameters.The prediction error of the model in porosity and permeability parameters is 3.51% and 15.17%,respectively,and the prediction accuracy is significantly improved.This method can effectively address the issues of parameters prediction in heterogeneous reservoir, and provide reference for the research combining artificial intelligence technology with physical models.

Key words: reservoir parameter prediction, transfer learning, SMOTE, random forest, neural network