Abstract
Drilling efficiency analysis typically relies on manual post-event analysis, which is subjective and arbitrary, failing to accurately reflect real-time field conditions in a timely manner. To enable real-time, accurate, and automatic identification of drilling conditions and improve drilling efficiency, the authors developed an intelligent identification model based on artificial neural networks. Using Pearson correlation coefficient for correlation analysis, eight drilling parameters from comprehensive logging data were selected as input for network training: well depth, bit position, hook height, hook load, weight on bit, rotary speed, torque, flow rate, and standpipe pressure. By comparing the performance of Long Short-Term Memory (LSTM) neural networks, BP, and CNN in real-time intelligent identification of drilling conditions in deep formations, it was found that LSTM outperformed the others. The LSTM model achieved a recognition accuracy of 97%, demonstrating its efficiency and reliability while providing important theoretical and technical support for effective drilling condition identification.