Prediction of Porosity and Permeability from Well-Logs Using Genetic Algorithm-Optimised Neural Networks
Accurate estimation of porosity and permeability from well-log data is essential for reliable reservoir characterization, yet the use of high-dimensional input features often increases model complexity and the risk of overfitting, particularly in data-limited settings. This study investigates the integration of a genetic algorithm (GA) with artificial neural networks (ANN) to identify the most informative well-log inputs for porosity and permeability prediction while reducing model complexity and improving model interpretability. Well-log data were preprocessed using linear interpolation for missing values and z-score normalization. A GA was employed within a wrapper framework to select optimal feature subsets, which were subsequently used to train multilayer perceptron ANN models. Model performance was evaluated using root mean square error (RMSE) and coefficient of determination (R2) for both porosity and permeability predictions. The GA consistently selected RHOB, NPHI, GR, and DT as the most influential input features, reflecting their physical relevance to pore structure, lithology, and rock mechanical properties. Compared with ANN models trained on the full feature set, the GA-ANN models achieved comparable predictive performance with reduced input dimensionality. For porosity prediction, validation RMSE reduced slightly from 0.0838 to 0.0821, while R2 improved from 0.9926 to 0.9929. For permeability prediction in log space, RMSE reduced from 0.0707 to 0.0657, whereas R2 improved from 0.9698 to 0.9740. Despite the modest RMSE deterioration, residual distributions became tighter and training behavior remained stable, indicating improved generalization and robustness. These results demonstrate that GA-based feature selection can simplify ANN models while maintaining competitive predictive capability, making the GA-ANN framework a robust and practical approach for petrophysical property estimation from well-log data in reservoir environments.