Abstract: The coal industry, as a pillar industry of China, has made outstanding contributions to the development of the national economy. However, the dust pollution generated during coal mining seriously affects the health and safety of underground workers. To address this issue, the traditional long-pressure short-extraction ventilation and dust removal system was improved based on the Coanda effect. The improved system was then subjected to intelligent parameter optimization using a convolutional neural network (CNN). Finite element modeling simulation and parameter optimization verification of the improved Coanda effect ventilation and dust removal system showed that the best dust removal effect was achieved at a pressure-extraction ratio of 2:3. Compared to the traditional long-pressure short-extraction ventilation system with the same pressure ratio, the improved system reduced the average dust concentration at the driver's position and the downwind side by 5.56% and 55.41%, respectively. An experimental platform with a scaled-down version of the improved Coanda effect ventilation and dust removal system was built to collect dust control parameters. Further, a neural network model was used to predict dust concentrations under different dust control parameters to achieve intelligent parameter optimization. The best dust removal effect control parameters for different initial dust concentrations were predicted and validated. It was found that the improved Coanda effect dust removal system achieved the highest dust removal efficiency when the initial dust concentration was 900 mg/m3 using the sixth control scheme, reducing the dust concentration at the driver's position and the downwind side by 80.35% and 87.42%, respectively. The results indicate that the improved Coanda effect dust removal system for the excavation face, after intelligent parameter control by the CNN, has high dust removal efficiency.