A hybrid machine learning optimization algorithm for multivariable pore pressure prediction

Abstract

Pore pressure is essential data in drilling design, and its accurate prediction is necessary to ensure drilling safety and improve drilling efficiency. Traditional methods for predicting pore pressure are limited when forming particular structures and lithology. In this paper, a machine learning algorithm and effective stress theorem are used to establish the transformation model between rock physical parameters and pore pressure. This study collects data from three wells. Well 1 had 881 data sets for model training, and Wells 2 and 3 had 538 and 464 data sets for model testing. In this paper, support vector machine (SVM), random forest (RF), extreme gradient boosting (XGB), and multilayer perceptron (MLP) are selected as the machine learning algorithms for pore pressure modeling. In addition, this paper uses the grey wolf optimization (GWO) algorithm, particle swarm optimization (PSO) algorithm, sparrow search algorithm (SSA), and bat algorithm (BA) to establish a hybrid machine learning optimization algorithm, and proposes an improved grey wolf optimization (IGWO) algorithm. The IGWO-MLP model obtained the minimum root mean square error (RMSE) by using the 5-fold cross-validation method for the training data. For the pore pressure data in Well 2 and Well 3, the coefficients of determination (R2) of SVM, RF, XGB, and MLP are 0.9930 and 0.9446, 0.9943 and 0.9472, 0.9945 and 0.9488, 0.9949 and 0.9574. MLP achieves optimal performance on both training and test data, and the MLP model shows a high degree of generalization. It indicates that the IGWO-MLP is an excellent predictor of pore pressure and can be used to predict pore pressure.