LSTM-Based Prediction Model for Tool Face State of Sliding Directional Drilling

For sliding directional drilling, because of the lack of quantitative control experience and the difficulty of physical modeling, the control parameters of the torsional pendulum system are artificially chosen according to the motion state of the tool surface, which leads to the effect of directional control depends on the artificial experience, thus the control efficiency is uneven, the problem of human resources cost is large and the efficiency is low. Aiming at more effective controlling of torsion pendulum in sliding directional operation, an intelligent prediction model for tool face state based on Long-Short Term Memory (LSTM) network is proposed to make more effective decisions for control parameters. Experiments show that the temporal sequence and nonlinear law in sliding directional operation can be well learned. Moreover, the effectiveness of the proposed prediction model is tested on several real oil drilling platforms. The research results have great potential applications in intelligent and automatic directional drilling process.