The shear wave(S-wave)velocity is a critical rock elastic parameter in shale reservoirs,especially for evaluating shale fracability.To effectively supplement S-wave velocity under the condition of no actual measuremen...The shear wave(S-wave)velocity is a critical rock elastic parameter in shale reservoirs,especially for evaluating shale fracability.To effectively supplement S-wave velocity under the condition of no actual measurement data,this paper proposes a physically-data driven method for the S-wave velocity prediction in shale reservoirs based on the class activation mapping(CAM)technique combined with a physically constrained two-dimensional Convolutional Neural Network(2D-CNN).High-sensitivity log curves related to S-wave velocity are selected as the basis from the data sensitivity analysis.Then,we establish a petrophysical model of complex multi-mineral components based on the petrophysical properties of porous medium and the Biot-Gassmann equation.This model can help reduce the dispersion effect and constrain the 2D-CNN.In deep learning,the 2D-CNN model is optimized using the Adam,and the class activation maps(CAMs)are obtained by replacing the fully connected layer with the global average pooling(GAP)layer,resulting in explainable results.The model is then applied to wells A,B1,and B2 in the southern Songliao Basin,China and compared with the unconstrained model and the petrophysical model.The results show higher prediction accuracy and generalization ability,as evidenced by correlation coefficients and relative errors of 0.98 and 2.14%,0.97 and 2.35%,0.96 and 2.89%in the three test wells,respectively.Finally,we present the defined C-factor as a means of evaluating the extent of concern regarding CAMs in regression problems.When the results of the petrophysical model are added to the 2D feature maps,the C-factor values are significantly increased,indicating that the focus of 2D-CNN can be significantly enhanced by incorporating the petrophysical model,thereby imposing physical constraints on the 2D-CNN.In addition,we establish the SHAP model,and the results of the petrophysical model have the highest average SHAP values across the three test wells.This helps to assist in proving the importance of constraints.展开更多
This study proposes a supervised learning method that does not rely on labels.We use variables associated with the label as indirect labels,and construct an indirect physics-constrained loss based on the physical mech...This study proposes a supervised learning method that does not rely on labels.We use variables associated with the label as indirect labels,and construct an indirect physics-constrained loss based on the physical mechanism to train the model.In the training process,the model prediction is mapped to the space of value that conforms to the physical mechanism through the projection matrix,and then the model is trained based on the indirect labels.The final prediction result of the model conforms to the physical mechanism between indirect label and label,and also meets the constraints of the indirect label.The present study also develops projection matrix normalization and prediction covariance analysis to ensure that the model can be fully trained.Finally,the effect of the physics-constrained indirect supervised learning is verified based on a well log generation problem.展开更多
基金supported by the National Natural Science Foundation of China(Nos.42374150,42374152)Natural Science Foundation of Shandong Province(ZR2020MD050).
文摘The shear wave(S-wave)velocity is a critical rock elastic parameter in shale reservoirs,especially for evaluating shale fracability.To effectively supplement S-wave velocity under the condition of no actual measurement data,this paper proposes a physically-data driven method for the S-wave velocity prediction in shale reservoirs based on the class activation mapping(CAM)technique combined with a physically constrained two-dimensional Convolutional Neural Network(2D-CNN).High-sensitivity log curves related to S-wave velocity are selected as the basis from the data sensitivity analysis.Then,we establish a petrophysical model of complex multi-mineral components based on the petrophysical properties of porous medium and the Biot-Gassmann equation.This model can help reduce the dispersion effect and constrain the 2D-CNN.In deep learning,the 2D-CNN model is optimized using the Adam,and the class activation maps(CAMs)are obtained by replacing the fully connected layer with the global average pooling(GAP)layer,resulting in explainable results.The model is then applied to wells A,B1,and B2 in the southern Songliao Basin,China and compared with the unconstrained model and the petrophysical model.The results show higher prediction accuracy and generalization ability,as evidenced by correlation coefficients and relative errors of 0.98 and 2.14%,0.97 and 2.35%,0.96 and 2.89%in the three test wells,respectively.Finally,we present the defined C-factor as a means of evaluating the extent of concern regarding CAMs in regression problems.When the results of the petrophysical model are added to the 2D feature maps,the C-factor values are significantly increased,indicating that the focus of 2D-CNN can be significantly enhanced by incorporating the petrophysical model,thereby imposing physical constraints on the 2D-CNN.In addition,we establish the SHAP model,and the results of the petrophysical model have the highest average SHAP values across the three test wells.This helps to assist in proving the importance of constraints.
基金partially funded by the National Natural Science Foundation of China (Grants 51520105005 and U1663208)
文摘This study proposes a supervised learning method that does not rely on labels.We use variables associated with the label as indirect labels,and construct an indirect physics-constrained loss based on the physical mechanism to train the model.In the training process,the model prediction is mapped to the space of value that conforms to the physical mechanism through the projection matrix,and then the model is trained based on the indirect labels.The final prediction result of the model conforms to the physical mechanism between indirect label and label,and also meets the constraints of the indirect label.The present study also develops projection matrix normalization and prediction covariance analysis to ensure that the model can be fully trained.Finally,the effect of the physics-constrained indirect supervised learning is verified based on a well log generation problem.
