Application of Recurrent Neural Network Long Short-Term Memory Model on Early Kick Detection

Abstract Long-short term memory [1] (LSTM) is an artificial Recurrent Neural Network (RNN) architecture capable of performing deep learning tasks. With the special feedback feature, the LSTM network is suitable for processing a sequence of data and making a sequence of predictions. It has been successfully applied to many disciplines such as speech recognition, language translation, time series forecasting, and anomaly detection. In this paper, the RNN-LSTM network is applied to real-time drilling data to study the complex dependencies between multiple drilling parameters and common kick indicators. A well-trained model will use the concept of the sliding window to continuously predict the unforeseen value of sensitive kick indicators. With proper analysis, the predicted result is helpful to detect kicks ahead of time. This paper also proposed a general workflow to easily visualize the prediction results. Compared with other time series prediction methods, the LSTM network has the advantages of more accurate multi-step prediction, more physical, and more flexible. The proposed LSTM network uses accelerated GPU computing, the fast computational speed makes both online and offline learning possible. It is concluded that this approach is capable of accurately predicting kick indicators under certain circumstances. It may provide innovative guidance for the application of the LSTM network in early kick detection and future study.