Bidirectional Recurrent Neural Network and Nondominated Sorting Genetic Algorithm II–Based Multiobjective Optimization Model for Drilling Parameters

Summary As oil and gas exploration targets deeper formations, the increased drilling depth and complexity reduce operational efficiency and extend drilling cycles. Traditional optimization methods, which rely on rate of penetration (ROP) and mechanical specific energy (MSE), often show significant prediction errors under complex downhole conditions. Therefore, more accurate and adaptive optimization methods are essential for efficient deep and ultradeep well operations. In this paper, we propose a downhole drilling parameter prediction model based on a bidirectional recurrent neural network (Bi-RNN). The model captures bidirectional temporal correlations between surface and downhole parameters. Through this approach, it accurately predicts the ROP and calculates MSE. The experimental results demonstrate exceptional performance on the test data set. The average relative error is only 2.99% for the ROP and 3.45% for the MSE calculated from the predicted values. To enhance real-time applicability, the model integrates the nondominated sorting genetic algorithm II (NSGA-II), creating a dynamic optimization system that provides Pareto-optimal solutions for surface drilling parameters. Field testing in the 5521–5859-m section of a deep well showed that the model increased ROP by 57.52% with only a 2.78% increase in MSE, effectively balancing drilling speed and energy efficiency. This optimization model serves as a practical tool for drilling engineers, supporting real-time decision-making to reduce drilling costs and improve drilling efficiency in deep and ultradeep well operations.