Artificial Neural Networks to Predict Lost Circulation Zones at Southern Iraq Oilfield

Abstract

Drilling soft and fragile areas such as (high permeable, cavernous, fractured, and sandy formations) have several problems. One of the most critical problems is the loss of drilling fluid into these formations in whole or part of the well. The loss of drilling fluid can lead to more significant and complex problems, such as pipe sticking, well kick, and closing the well. The drilling muds are relatively expensive, especially oil-based mud or those that contain special additives, so it is not economically beneficial to waste and lose these muds. Artificial neural networks (ANN) can predict drilling fluid losses before they occur based on drilling parameters data and drilling fluid properties of wells effected by lost circulation problems located in the same area. This paper developed two artificial neural network models to predict drilling fluid losses in the Dammam formation- Rumaila oil field in southern Iraq. The two models have the same topology and structure. The first model used the early stopping technique to stop the training when we get the global minimum and the second model used specific epochs to complete the training. The models could predict various types of losses with high accuracy. The accuracy of implementing R2 for the first and second models was 0.9302 and 0.9493, respectively. The early stopping technique lead to obtain a model with acceptable accuracy in a short time without relying on a specific number of epochs.