Real-time prediction of logging parameters during the drilling process using an attention-based Seq2Seq model

In recent years, there has been a notable upsurge within the drilling industry regarding the construction of machine learning models that leverage logging parameters to augment decision-making processes. When building these models, the logging parameters are usually assumed to be fully accessible. However, it may not always hold true because certain logging parameters will be inaccessible or unavailable in real-time drilling operations, which greatly hinders the practical implementation of these models. Aiming at addressing the logging parameters' accessibility issue, this paper proposes an innovative attention-based sequence-to-sequence (Seq2-Seq) model to enable real-time predictions. The proposed model integrates the Seq2Seq model, bidirectional Gated Recurrent Unit (GRU) model, and attention mechanism to enable multi-step ahead prediction. Further-more, dynamic updates using the sliding window method with intermittent frequency are employed to enhance the model's performance. Meanwhile, hyperparameter tuning is automatically performed using an open-source toolbox to boost the model training process. Finally, the efficacy of the model is evaluated by predicting four commonly utilized logging parameters (surface torque, standpipe pressure, mud pit volume, and Gamma ray) based on two field drilling datasets. The comprehensive analysis demonstrates promising results, with mean absolute percentage errors below 5% across the datasets. Moreover, the attention-based Seq2Seq model surpasses base models (the vanilla GRU model and the Seq2Seq model) due to the inherent Seq2Seq structure, bidirectional architecture, and attention layer. This paper offers a practical approach to bridge the gap between the devel-opment of theoretical machine learning models and their implementation in real-world drilling processes.