This study investigates the impact of varying traffic volumes, equipment rates, and warning locations on traffic flow in foggy highway sections under a vehicle–road collaborative warning environment using simulation methods. A double hidden layer BPNN model is employed for modeling and prediction, achieving high fitting and forecasting accuracy by selecting appropriate hidden layer nodes and activation functions. The findings indicate that higher equipment rates lead to improved overall efficiency metrics, with the optimal warning distance varying under different traffic conditions. In free-flow conditions, the optimal warning distance ranges from 1000 to 3200 m, while in saturated and oversaturated conditions, it ranges from 1200 to 1800 m and 200 m to 600 m, respectively, with minimal impact from increasing traffic volumes. The results demonstrate that this model effectively enhances warning service quality in foggy highway sections and provides valuable insights for similar modeling and prediction challenges.