Abstract
Accurate prediction of rate of penetration (ROP) plays a pivotal role in optimizing pre-drilling resource allocation, developing cost-effective drilling strategies, and enhancing operational efficiency. This study proposes a novel ROP prediction model integrating particle swarm optimization (PSO) with a momentum-adaptive backpropagation (BP) neural network. Utilizing drilling data, we first conducted comprehensive correlation analysis between drilling parameters and performed theoretical validation through conventional ROP equations. Critical input parameters were systematically selected, including three engineering indices, two hydraulic characteristics, and six lithological properties. Twelve hundred data points were selected as the sample datasets, 80% for the training, 20% for the testing. The developed algorithm incorporates two heuristic optimization techniques: (1) momentum acceleration for gradient descent stabilization, and (2) adaptive learning rate adjustment for enhanced training efficiency, predictive accuracy, and model generalizability. Analytical results reveal significant correlations coefficient of greater than 0.5 between ROP and the identified parameters. Through comparative experiments with field data, the optimized PSO-BP model demonstrated superior performance over standard BP and genetic algorithm-enhanced BP (GA-BP) variants, achieving MAE of 0.30 m/h, MAPE of 11.35%, and R² of 0.93. This data-driven approach establishes a reliable framework for ROP prediction in well-characterized formations with comprehensive logging datasets.