Intelligent recognition method for drilling conditions of deep formation based on LSTM neural network

Deep formation drilling is characterized by high investment and risk. Real-time identification of drilling conditions is fundamental and prerequisite for improving drilling efficiency and reducing complex incidents. Traditionally, drilling operations relied primarily on manual post-analysis, which suffers from subjectivity and arbitrariness. This approach fails to promptly and accurately reflect actual field conditions, making it difficult to ensure timeliness and accuracy. Aiming to achieve real-time and accurate automatic identification of drilling conditions and enhance drilling efficiency, this study innovatively introduced machine learning into the entire process of identifying working conditions in deep formation drilling. By comprehensively analyzing the long-term time-series characteristics of drilling data, the authors established a real-time intelligent identification model based on a Long Short-Term Memory (LSTM) neural network. Eight drilling parameters, such as bit depth, well depth, hook height, weight on bit, weight on hook, torque, rotary speed, and standpipe pressure, were selected as input values. A 16-hidden-layer×80-node LSTM neural network model was developed. The model achieved a remarkable accuracy rate of up to 93.62% on the training set and 93.05% on the testing set. It fully demonstrated its high efficiency and reliability. LSTM attained the highest values across these metrics, with scores of 0.93, 0.91, and 0.88, compared with CNN, RNN, and SVM. Application results indicate that the model performs optimally in identifying drilling conditions. It enables real-time intelligent recognition of drilling conditions, improves drilling efficiency, and aligns with the requirements for digital and intelligent development in oilfields. This provides important theoretical and technical support for the efficient identification of drilling working conditions.