Abstract: Abstract:To address the challenge of accurately monitoring the average air velocity due to uneven airflow distribution in tunneling drives, and to advance the development of intelligent mine ventilation, this study proposes a BP neural network prediction model optimized by an Improved Aquila Optimizer (IAO). First, computational fluid dynamics (CFD) numerical simulations were employed to systematically obtain flow field data under various conditions of roadway dimensions, duct sizes, and air volumes, resulting in a dataset containing 3032 samples. Based on this dataset, the predictive performance of three algorithms—Support Vector Machine (SVM), Random Forest (RF), and the BP neural network—was compared, identifying the BP neural network as the optimal base model. To further enhance prediction accuracy and stability, multiple metaheuristic optimization algorithms were introduced and compared, ultimately selecting the Improved Aquila Optimizer (IAO) to optimize the weights and thresholds of the BP neural network, thereby establishing the IAO-BP prediction model. Validation results from numerical simulations show that the model achieved a mean relative error of 5.11%, with relative errors consistently below 9%. Field test validation results demonstrated that the model achieved mean relative errors of 6.95% and 7.52% in two actual tunneling roadways, respectively. The study confirms that the IAO-BP neural network model can accurately and reliably predict the average air velocity in a roadway cross-section based on a single-point wind speed measurement within that section, providing an effective technical tool for real-time monitoring and intelligent regulation of mine ventilation systems.