A machine learning approach to predict geomechanical properties of rocks from well logs

The estimation of geomechanical properties such as Young's modulus from well logs is a crucial task in the field of geomechanical engineering. Traditional methods for estimating these properties can be time-consuming and may not always provide accurate results. In recent years, machine learning techniques have been increasingly used for this purpose, providing a more efficient and accurate alternative. This paper focuses on the estimation of Young's modulus from well logs using machine learning techniques. The use of well logs as input data for machine learning models has several advantages, including the ability to handle large datasets, the identification of key input features, and the creation of predictive maps. Well logs from well 15/9_F_11A of Volve Field from a depth range 3575-3720 m are used for training the various ML algorithms mainly the support vector regression, linear regression, and decision tree and also for generating the synthetic Young's modulus. The well 15/9_F_1A of Volve Field is used for predicting Young's modulus. From the train-test results, decision tree has shown a test score of 0.97. The score metrics are calculated, and the decision tree has a MAPE of 0.034%, which is the least among the other ML algorithms used for prediction. The actual and predicted values of Young's modulus have shown good precision results. The model shows greater efficiency to analyse large amounts of well log data and presents a cost-effective solution to predict Young's modulus in wells which contain less information due to financial or technical reasons. Also, the stratigraphy of the training well and testing well is not completely similar, but the model has generated an R2 of 0.973 on the test well. Finally, the use of machine learning for estimating Young's modulus from well logs is proved to be a promising approach that can provide valuable insights and results for geomechanical engineering applications.