Advanced Porosity Prediction in Heterogeneous Oil Reservoirs: Using Novel Machine Learning and Deep Learning Techniques

The estimation of porosity holds significant importance within the oil and natural gas sectors as it plays a crucial role in the accurate assessment of reservoir characteristics. Hence, it becomes paramount to predict porosity optimally in order to minimize the need for logging and core tests. The primary contribution of this study lies in the examination of the effectiveness of five machine learning (ML) techniques, along with Artificial Neural Networks (ANN), for predicting porosity. This investigation uses data from X-Iraqi heterogeneous oil reservoirs, which consists of 311 data points, with core porosity as the reference output. This work was conducted using the Python programming language using the Spyder platform, coupled with the utilization of Interactive Petrophysics (IP) software. Prior to model training, data processing was performed, including data cleaning, normalization, and division into training (70%) and testing (30%) groups. The developed models of porosity prediction are based on six input datasets consisting of well-known logs: gamma ray (GR), neutron porosity (NPHI), density (RHOB), spontaneous potential (SP), micro-log lithology (MLL), and sonic (DT). Statistical analysis and visualization approaches were utilized to interpret the efficacy of the anticipated models. The findings revealed that Artificial Neural Networks (ANN) and Decision Tree Model (DTM) generated superior predictive capabilities compared to other ML approaches evaluated. ANN attained the highest average coefficient of determination (R2) of 0.87 as well as the lowest root mean square error (RMSE) of 0.024. Similarly, DTM achieved an average R2 of 0.85 and RMSE of 0.025. These results highlight the effectiveness of both ANN and DTM in accurately estimating porosity in heterogeneous reservoirs. The statistical analysis validated the significant enhancement obtained by ANN and DTM in capturing the complex spatial variations and intricate geological features of the reservoirs, surpassing the precision of traditional porosity prediction methods. The visualization graphs further verified the findings by showing the improved predictive performance of ANN in accurately depicting the reservoir porosity distribution. In contrast, the traditional methods failed to match the core porosity measurements appropriately. Based on the evaluations, it was discovered that K-Nearest Neighbors (KNN), demonstrated efficient performance in the training phase. However, it failed to satisfy the acceptability criteria during testing. Consequently, these findings emphasize the revolutionary possibilities of ANN and DTM, for accurately predicting porosity in heterogeneous reservoirs. To study the robustness and sensitivity of the proposed models, sensitivity analysis was carried out, offering insights into the influence of input parameters on the porosity prediction outputs. Based on the analysis, it was determined that neutron porosity (NPHI) had the highest impact on porosity prediction. As a result, the outcomes of this work have significant consequences for reservoir characterization and development planning, enabling more informed decision-making and optimized hydrocarbon recovery strategies.