.png){kind=icon}
.png){kind=icon}
Return
A Knowledge-Guided Data-Driven Method for Predicting Reservoir Parameters
.png){kind=icon}
PRE
AI
DOI：10.30632/PJV67N2-2026a7.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Focusing on the Sulige area of the Ordos Basin in China, this paper integrates a data-driven approach with petrophysical models to achieve high-precision prediction of key reservoir parameters, including porosity, water saturation, and permeability. An augmented data set is constructed using conventional logging data combined with computed results from Archie's formula, the Timur formula calibrated with regional statistics, and the neutron porosity logging interpretation model. Core experimental data are used as training labels for evaluation. After data preprocessing, multiple mainstream machine-learning models are comprehensively evaluated, revealing that the deep neural network (DNN) delivers the best overall performance. To further enhance feature extraction and generalization ability, this paper innovatively introduces dilated convolutional neural networks (DCDNN) into the DNN architecture, constructing a DCDNN model. Ablation experiments confirm that this strategy significantly improves the model's representational ability. To boost model robustness and prediction accuracy, the multi-player dynamic game (MPDG) algorithm and Bayesian optimization are applied for hyperparameter tuning of the DCDNN. Experimental results demonstrate a substantial improvement in the optimized model's predictive performance. Finally, based on ensemble learning theory, an expert committee-based decision-making mechanism is established, and an optimal prediction model is selected using multiple comprehensive metrics as evaluation criteria. Compared with the standalone DNN model, the ensemble model reduces the mean absolute error in porosity, water saturation, and permeability predictions by 1.066, 12.711, and 1,661, respectively. The R2 values improve by 3.33, 2.95, and 23.9%, while the relative percent difference values improve by 1.32, 0.48, and 0.87, respectively. Applied to a blind well, the model achieves excellent predictive performance, providing reliable technical support for geological structure analysis and resource assessment in the region.
Keywords:
MACHINE
REGRESSION
ALGORITHM
POROSITY
PERMEABILITY