Título: Hybrid computational model to predict Total Organic Carbon
Autores: Clovis Antonio da Silva, Juliana da Costa Cabral, Grazione de Souza Boy & Camila Martins Saporetti
Resumo: One important metric for assessing the amount of organic matter in source rocks is the Total Organic Carbon (TOC) content of rock samples. On the other hand, calculating TOC requires the collecting and analysis of samples from numerous well intervals in the source rock formations, which is a very resource-intensive process. Researchers have looked for creative ways to simplify TOC estimation in order to overcome this difficulty. Among these, machine learning techniques have demonstrated potential as an alternative to traditional well log analysis and stratigraphic study. This work focuses on automating TOC estimate utilizing advanced machine learning techniques enhanced by a hybrid methodology to improve the models’ accuracy and adaptability. Four metaheuristic algorithms—Arithmetic Optimization Algorithm, Coronavirus Herd Immunity Optimizer, Differential Evolution, and Particle Swarm Optimization—were used to optimize the machine learning models. Four machine learning techniques—Decision Tree, Extreme Learning Machine, Gradient Boosting, and K-Nearest Neighbors—incorporated these metaheuristics. Core samples from the Shahejie Formation, Dongying Depression, Bohai Bay, China were used to evaluate the hybrid technique. The findings show that a hybrid approach combined with machine learning models produces extremely accurate and flexible models for TOC prediction. Regardless of the metaheuristic that was applied to direct the model selection process, the optimized Extreme Learning Machine produced the best performance metrics. These results demonstrate how hybrid models can improve exploratory geological research by providing a more accurate and efficient way to estimate TOC in source rocks.
Palavras-chave: total organic carbon; machine learning; metaheuristics; hybrid models.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1141185
Artigo em PDF: CBIC_2025_paper1141185.pdf
Arquivo BibTeX:
CBIC_2025_1141185.bib