| Abstract |
Directional drilling in naturally fractured geothermal reservoirs is a challenging task due to unexpected changes in inclination and azimuth of the well axis, which cause inefficient weight on bit transfer, decrease in penetration rate, increasing the risk of stuck pipe and problems in while running casings. To predict the sudden changes in inclination while drilling, a back propagation, feed forwarded multi layered artificial neural network (ANN) model, which uses drilling data collected from 12 J-type directionally drilled geothermal wells from Büyük Menderes Graben, Turkey was developed. The training data consisted of 7600 individual drilling data. During the training process, effects of each drilling parameter on inclination were investigated with different scenarios for different hole sizes. Moreover, inclination predictions were carried out for a field case in which kick off point to the target depth with 30 meters survey intervals and results were compared. It has been found that developed ANN model provided satisfactory results based on the mean-square-error value which was measured to check accuracy and quality of each training. It has been found out that as weight on bit (WOB), bit revolution per minute and stand pipe pressure increase, inclination increases. On the other hand, increment in flow rate leads to drop in inclination. The results indicated that, total flow area, bit hardness and WOB have the highest impact on network data compared to other drilling parameters. |