Artificial neural network prediction of wellbore stability in offshore shallow formations

Abstract

Listen

Translate article

Wellbore instability is one of the most critical challenges during drilling, which may result in complex problems such as stuck pipe, high torque and mud loss, impeding the drilling progress and increasing the cost of drilling operations. Offshore shallow formations are characterized by weak consolidation and low strength, causing serious wellbore collapse frequently. The traditional physics-based wellbore instability models attempt to predict the risk of failure of the wellbore for given drilling fluid densities and provide optimal fluid density for safe drilling. However, these models generally involve quite a few empirical coefficients, determination of which heavily rely on the field experiences of the engineers, and thus the results may vary greatly from person to person. In this study, an artificial neural network (ANN) model has been established for wellbore stability prediction, aiming to reduce the subjectivity while mapping the drilling performance of neighboring wells into the predicted results. Eleven factors that may affect wellbore stability have been identified from the prevailing physics-based model and used as the input of the ANN model, while the wellbore enlargement rate (WER) is used as the output to quantify the wellbore stability performance. The model has been trained using data from 5 wells drilled in the offshore shallow formations of the Bohai Bay Basin in east China and then used to predict the WER of another well in the same region. The results show that the mean absolute percentage error is 6.113%, and at most well depths the predicted wellbore diameter deviate from the measured values not more than 5%. Comparison of the ANN model and some other machine learning models, including random forest model, decision trees model, linear regression model, was conducted, which demonstrated the best performance of the ANN model in terms of the predicted wellbore diameter profile. Finally, the potential application of the ANN model in optimizing mud weight has been illustrated through the predicted wellbore diameter profiles for different mud weights. It is worth noting that the method presented in this paper is of physics and data-driven nature and provides a new research insight for wellbore stability analysis.