In offshore drilling, operators perform several operations to facilitate drilling. For some of these operations, the operators must rely on quick decision-making by inspecting physical variables in real time. However, uncertainties arise when the relationship between these variables changes due to changes in the physical environment. As the drilling operations are influenced by combinations of complex physical principles, these changing relationships make the decision-making more challenging. If poor decisions are made, it could result in loss of well control and, in the worst case, a blowout. This is the underlying motivation for this paper.
In order to capture the uncertainty of the changing relationship between the variables, an approach is developed to support the operators in decision-making that leads to the desired drilling regime. In this approach, we train a machine-learning model to learn the relationship between two variables using a data sample for a given time window. The trained model is then used to estimate and predict these variables. Our focus is on modelling the dynamic relationship between two key variables, i.e. an independent control variable and a dependent response variable, using an artificial neural network (ANN). The uncertainty of the model prediction is addressed by retraining the ANN model (so-called re-enforcement learning), which is achieved whenever the value of the system response exceeds the set threshold value. In this way, our approach not only predicts the system response for a given input but also implicitly accounts for the uncertainty of the changing physical environment that may increase the model error.
An example has been presented to demonstrate the use of this approach in an oil and gas well drilling simulator, where the operator controls the flow rate of drilling fluid to achieve a target pressure. We find that our approach satisfactorily predicts the input flow rate to achieve a target output pressure in the simulated system but should also be complemented with consideration of the assumptions and simplifications on which it is built.