Feature Cross and Time-Series Memory Network for Intelligent Rate of Penetration Optimization in Offshore Drilling

The rate of penetration (ROP) is a crucial indicator for evaluating efficiency in offshore drilling engineering. Traditional ROP optimization relies on expert systems requiring manual parameter adjustments by specialists. In this paper, we propose a deep learning-based intelligent ROP optimization model named the Feature Cross and Time-series Memory Network (FCTMN). The model designs a predict-optimize workflow that integrates an ROP prediction model with an ROP optimization model to automatically generate recommended parameters. Experiments conducted on real-world datasets demonstrate that FCTMN can provide reliable parameter recommendations and significantly improves drilling speed.