A hybrid machine learning framework for accurate petrophysical prediction: application to formation resistivity in the Nubian reservoir
Year: 2026
Type: article
Abstract: Utilizing Machine Learning (ML) approaches to predict petrophysical parameters is becoming increasingly common due to their effectiveness in identifying non-linear relationships. Formation resistivity (RT) is a crucial petrophysical parameter that significantly impacts geophysical and geological applications; however, although ML algorithms have shown effectiveness in forecasting RT, several limitations have been identified in numerous investigations, particularly in handling the complex relationships within heterogeneous reservoirs. This study proposes two innovative hybrid ML frameworks for RT prediction using data from the Nubian reservoir in Libya. The primary hybrid model (ELM-RBF) integrates an Extreme Learning Machine (ELM) with a Radial Basis Function (RBF) network, while the secondary hybrid model (MLP-GPR) amalgamates a Multi-layer Perceptron (MLP) with a Gaussian Process Regressor (GPR). All algorithms are meticulously optimized via Grid Search Optimization (GSO). The models were evaluated against the corresponding standalone models. Prior to model development, a comprehensive data preprocessing workflow was implemented, incorporating a novel denoising process to enhance data quality. The proposed hybrid models have superior predictive accuracy and consistency compared to the Standalone models. Improved generalization performance and reduced overfitting were confirmed through multiple evaluation metrics, including the Generalization Gap (GG), Generalization Error Ratio (GER), and Normalized Root Mean Square Error (NRMSE). Ensemble-based uncertainty quantification further indicates lower predictive uncertainty and enhanced robustness of the hybrid models. Additionally, a novel empirical equation for RT prediction was developed using Gene Expression Programming (GEP), providing a reference for estimating RT in the study area and adjacent regions where implementation of hybrid models may be challenging.
Source: Frontiers in Earth Science
Authors: Mohammed A. Amir(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Hamzah S. Amir(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Mansour Almazroui(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Riaz Ali(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Rizwan Niaz(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Irfan Ur Rashid(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing), Ahmad Salah(Conceptualization, Formal analysis, Methodology, Visualization, Writing – original draft), Mohamed Elhag(Conceptualization, Formal analysis, Methodology, Writing – original draft, Writing – review & editing)
Institutions: University of Benghazi, University of East Anglia, King Abdulaziz University, Shenzhen University, Yunnan Normal University, Government of Pakistan, Mediterranean Agronomic Institute of Chania, University of Thessaly, Chinese Academy of Sciences, German University of Technology, Aerospace Information Research Institute
Language: English
Subfield: Geology
Field: Earth and Planetary Sciences
Domain: Physical Sciences
Keywords: petrophysical properties, hybrid machine learning, extreme learning machine, radial basis function neural network, multilayer perceptron, Gaussian process regression, data denoising, generalization gap, uncertainty quantification, gene expression programming, empirical prediction equations, heterogeneous reservoirs
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7235563831
