Abstract: To address the problem that there are relatively few studies on the airflow field distribution and the position of average wind speed points in heading roadways at the working face (heading roadways) affected by obstacles, a rectangular heading roadway containing obstacles such as a roadheader and a forced-air duct was taken as the research object, and an average wind speed prediction model for heading roadways based on a BP neural network optimized by the Improved Aquila Optimizer (IAO), namely the IAO-BP model, was proposed. A physical model of the excavation roadway was constructed, and a wind speed distribution dataset under obstacle disturbance was obtained. Twenty-four three-dimensional geometric models of heading roadways covering six roadway section specifications and four air duct diameters were established, and steady-state turbulence simulations under 192 working conditions were completed using the standard − turbulence model and the SIMPLEC pressure-velocity coupling algorithm. The 9-point wind speed measurement method was used to monitor the wind speed variation across roadway cross-sections, and sensors were arranged at the center point and edge points. To effectively evaluate the applicability of different models in wind speed prediction for heading roadways, three typical machine learning algorithms, namely Support Vector Machine (SVM), Random Forest (RF), and BP neural network, were selected for comparison, and the BP-based average wind speed prediction model for heading roadways (BP model) was selected as the optimal basic model. To fully explore the mapping potential of the BP model in wind speed field data, the IAO algorithm was introduced to optimize the BP model, and the IAO-BP model was constructed. The results showed that the mean squared error of the IAO-BP model decreased to 0.1 and the coefficient of determination increased to 0.98, achieving efficient global optimization and high-precision mapping on complex error surfaces. The field test results showed that the average relative errors of the IAO-BP model in two heading roadways were 6.95% and 7.52%, respectively. The prediction accuracy meets the engineering requirements of mine ventilation monitoring, and the model can achieve accurate and reliable prediction of average wind speed based on the wind speed at any single point within the cross-section.