A deep learning joint decision model of variable speed limit and on-ramp control for traffic flow control was proposed to solve the problems of mutual isolation, feedback lag and insufficient coordination between the main line speed limit regulation and on-ramp control in the bottleneck area of expressway. The model integrates roadside sensing data, short-term state prediction and closed-loop feedback update mechanism, and synchronously generates the main line speed limit level and ramp release strategy in a unified state space to achieve collaborative suppression of speed attenuation, density accumulation and queue diffusion. Experiments were carried out based on 12.4 km continuous bottleneck road, two on-ramps and 16 weeks of traffic data. The results show that compared with the fixed speed limit strategy, the proposed method increases the average speed of the main line by 10.4%, increases the traffic per unit hour by 10.5%, reduces the congestion duration by 50.6%, and decreases the average queue length of the ramp by 46.4%. The research shows that the joint control framework driven by deep learning can improve the real-time performance, stability and overall efficiency of highway traffic regulation.
