Prediction of Bottom-Hole Pressure Differential During Tripping Operations Using Artificial Neural Networks (ANN)

Abstract

Tripping in or out drill string/casing with a certain speed from the wellbore will result in downhole pressure surges. These surges could result in well integrity or well control problems which can be avoided if pressure imbalances are predicted before this operation engaged. To predict these pressure imbalances, number of analytical models have been developed but require time-consuming cumbersome numerical analysis. In this paper, an intelligent model (ANN) is developed which can predict the surge pressure under varying rheological and geometrical parameters. ANN is developed with six neurons in input layer representing six input parameters (pipe velocity, PV, YP, diameter of hole, outer diameter of pipe and mud weight) and one neuron in output layer which represents surge pressure. Now, to find the most optimum neural network structure (number of hidden layer and neurons), total 108 ANN configuration is trained and tested. Performance analysis on these configurations indicates network structure with two hidden layers including ten and 16 neurons in first and second layer, respectively, as the most optimum. Since the selected model is complex, another trained model with one hidden layer containing 14 nodes can be considered due to its satisfactory prediction result. The trained intelligent model can be utilized when tripping operation is carried out in low-pressure margin wells where repetitive calculation of surge/swab pressure is required.