Real-time prediction of in-situ stresses while drilling using surface drilling parameters from gas reservoir

Subsurface planning and modeling are highly dependent on the accurate determination of in-situ stresses. For example, well instability, and hydraulic fracture operations highly depend on the values of in-situ stress. In-situ stress setting can be described in terms of three orthogonal components; overburden stress (σ_v) and minimum (σ_h) and the maximum (σ_H) horizontal stresses. σ_v can be determined using the density logs. σ_h and σ_H can be estimated from borehole injection tests or theoretical finite elements methods. However, these methods are complex, expensive, or need unavailable tectonic stress data. Therefore, The objective of this study is to introduce the application of machine learning (ML) techniques to predict σ_h and σ_H from the surface drilling data. The drilling data values vary in response to the in-situ stress. This work demonstrates how ML techniques can predict the minimum and maximum horizontal stresses using drilling data while drilling that were collected from a gas reservoir. Random forest (RF), functional network (FN), and adaptive neuro-fuzzy inference system (ANFIS) were implemented to predict σ_h and σ_H during drilling using measured surface drilling parameters. A dataset of 2573 points from two wells (Well-1 and Well-2) was used to build the different ML models. The data from Well-1 were used to train and test the model, and Well-2 data were then used to validate the developed models. The three ML models accurately predicted the σ_h and σ_H based on the drilling data. The three models showed similar results for σ_h prediction with a correlation coefficient (R) values greater than 0.96 and an average absolute percentage error (AAPE) less than 0.53%. Comparing the results of RF, ANFIS, and FN models for σ_H prediction showed that RF outperforms the other two models with R values for the training dataset of 0.99 compared to 0.94 and 0.87 for ANFIS, and FN, respectively. The RF, ANFIS, and FN models were able to capture the σ_h and σ_H trends with depth for the validation data. These results show the ML capabilities for real-time predicting σ_h and σ_H while drilling from the surface drilling parameters without additional costs.