Development of artificial neural network for predicting drill pipe sticking

Stuck piping is a common problem with tremendous impact on drilling efficiency and costs in oil industry. Prediction of stuck at the stage of designing and in the process of drilling wells, minimizes the risk of the occurrence of sticking, due to the choice of the optimal method of prevention for specific geological and technical conditions. The article is devoted to the development of an artificial neural network for prediction of sticking of drill pipe columns. The paper provides a literature review of existing methods of prediction of sticks. As input data elements are used important and generalizing factors influencing the emergence of all types of sticks, which allows predicting all types of sticks of drill pipe columns. In order to increase the susceptibility of the input data to the neural network, the data elements are transformed and normalized. The type and architecture of the network, as well as its hyperparameters, are chosen by the experimental method. Assessment of the quality of the network is made by the method of k-fold cross-validation. In order to find the optimal combination of activation function with various optimizers, experimental research is carried out with the analysis of their results. The experiments were implemented in the Python programming language with KERAS, TensorFlow and Matplotlib library packages, as well as in the cloud platform Colaboratory from Google. A distinctive feature of the proposed method is that the resulting forecasting model can be easily adapted to new data, which often occurs when drilling wells in new fields.