Abstract: In response to the issues of complex coupling among key parameters, reliance on empirical regulation, and limited dust reduction efficiency in the forced-exhaust ventilation and dust control system for fully mechanized excavation faces in coal mines, a parameter collaborative optimization method integrating BP neural network and policy gradient reinforcement learning (BP+PG) is proposed. Orthogonal experiments were conducted to obtain 82 sets of working condition data covering seven factors, including forcing airflow rate and exhaust airflow rate. A nonlinear mapping model between system parameters and the normalized dust concentration (A?) was constructed. Subsequently, a policy gradient algorithm was introduced to endow the model with adaptive parameter adjustment capability, and it was coupled with the PSO algorithm to perform global optimization on three key parameters: exhaust airflow rate, radial-to-axial airflow ratio, and the distance from the dust control device to the working face. Experimental results show that the BP+PG model achieves a mean absolute percentage error of only 3.40% and a coefficient of determination (R2) of 0.97 in prediction. After applying the optimal parameter combination obtained by PSO (450 m3/min, 1.5, 16 m), the measured dust concentration at the driver's position decreased by 63.2%, with a significant improvement in dust reduction efficiency throughout the entire roadway. This method provides a theoretical basis and a technical solution for the intelligent optimization of parameters in forced-exhaust ventilation and dust control systems.