Applied machine learning-based models for determining the magnitude of pore pressure and minimum horizontal stress

Determination of pore pressure and minimum horizontal stress are the first steps in geomechanical studies, and they are critical for hydrocarbon exploration, drilling, and production operations. The right quantity of pore pressure and minimum horizontal stress can enhance the quality of decisions taken and the economics of the oil and gas industry operations. Recent studies indicate that machine learning methods can be suitable for estimating the pore pressure and minimum horizontal stress, but it still needs to analyze different datasets, especially complex data to emphasize the accuracy of the results. Thus, the purpose of this article is to provide a robust machine-learning method to estimate the pore pressure and minimum horizontal stress using well-log data. The machine learning method includes multi-layer perceptron (MLP) and radial basis function (RBF). The performance of the MLP and RBF is dependent on the number of hidden layers and the number of neurons. To overcome this problem, the Genetic Algorithm (GA) is used to compute the number of hidden layers and number of neurons in each layer. A dataset of nine wells with 75,123 data points on the Volve oil field with various reservoir rocks is used to develop the predicted models. The results show the reliability of the machine learning methods to estimate the pore pressure and minimum horizontal stress from well-log data. It shows similar prediction accuracy when using unseen data. The comparison of MLP and RBF models presents that the MLP model for predicting and estimating pore pressure and minimum horizontal stress has higher accuracy than the RBF model.