Case Study: Neural Network Implementation in Ensemble Machine Learning for Well Log Estimation, Case Applied in Campos Basin

Abstract

Summary Several activities in geosciences are supported by hard data, which are represented by trustworthy information. However, not all wells offer basic logs such as sonic and density. This kind of information is significant for characterization in reservoir geophysics. This case study proposes a combination of Multilayer Perceptron (MLP) tools that constitute a type of Artificial Neural Network (ANN) and the Ensemble Machine Learning (EML) technique, in the prediction of missing or imputation log data based on the dataset of the Campos Basin. Such machine learning tools are considered robust, fast, and low cost, widely used in several areas. The study explores the combination of MLP and EML in the development of the learning algorithm. The use of MLP was “tuned” with optimal hyperparameters through GridSearch and the EML built through the Voting Estimator technique in a weighted way through the Scikit-learn library. It’s selected well logs like sonic, density, porosity, among other information for training. The velocity profile was selected as the prediction target. The best calculation parameters and errors of ensemble machine learners were generated, and thus, to analyze the generalizability of the algorithms. And finally, the EML Results were compared with the test samples.