Early Kick Detection Using Real Time Data Analysis with Dynamic Neural Network: A Case Study in Iranian Oil Fields

Abstract

Abstract One of the important challenges in drilling a well is kick detection. In fact, ignoring the signs of a kick lead us to complicated problems in killing the well and consequently, if the kick could not be controlled a blowout may occur. Therefore, considering the early detective parameters seems crucial. Kick happens because some parameters like tank volumes are changed when the time passes, so static neural network cannot predict this occurrence; therefore, using dynamic neural network, which incorporate the previous data could predict the system manner better than the static one. In this paper, a dynamic neural network model is presented and good results with acceptable accuracy are shown. The model was trained with some Iranian onshore oil wells that kick was encountered in them. The kicks were predicted longer before it could have been detected by the drilling crew.