A Novel Hybrid Machine Learning Approach and Basin Modeling for Thermal Maturity Estimation of Source Rocks in Mandawa Basin, East Africa

Abstract

Basin modeling and thermal maturity estimation are crucial for understanding sedimentary basin evolution and hydrocarbon potential. Assessing thermal maturity in the oil and gas industry is vital during exploration. With artificial intelligence advancements, more accurate evaluation of hydrocarbon source rocks and efficient thermal maturity estimation are possible. This study employed 1D basin modeling using PetroMod and a novel hybrid group method of data handling (GMDH) neural network optimized by a differential evolution (DE) algorithm to estimate thermal maturity (Tmax) and assess kerogen type in Triassic–Jurassic source rocks of the Mandawa Basin, Tanzania. The GMDH–DE addresses the limitations of conventional methods by offering a data-driven approach that reduces computational time, overcomes overfitting, and improves accuracy. The 1D thermal maturity basin modeling suggests that the Mbuo source rocks reached the gas–oil window in late Triassic times and began expulsion in the early Jurassic while located in an immature-to-mature zone. The GMDH–DE model effectively estimated Tmax with high coefficient of determination (R² = 0.9946), low root mean square error (RMSE = 0.004), and mean absolute error (MAE = 0.006) during training. When tested on unseen data, the GMDH–DE model yielded an R² of 0.9703, RMSE of 0.017, and MAE of 0.025. Moreover, GMDH–DE reduced the computational time by 94% during training and 87% during testing. The results demonstrated the model’s exceptional reliability compared to the benchmark methods such as artificial neural network–particle swarm optimization and principal component analysis coupled with artificial neural network. The GMDH–DE Tmax model offers a unique and independent approach for rapid real-time determination of Tmax values in organic matter, promoting efficient resource assessment in oil and gas exploration.