To address the instability issues of heterogeneous traffic flow in highway lane reduction sections, a comprehensive safety potential field model integrating road environment and vehicle interaction risks is established. Spatial correction factors and temporal correction factors are introduced to quantify the differentiated risks between Connected and Automated Vehicles (CAVs) and Human-Driven Vehicles (HDVs). A distributed Model Predictive Control (MPC) framework centered on string stability is constructed to achieve coordinated optimization control of heterogeneous traffic flow. Theoretical analysis derives the critical CAV penetration rate for maintaining traffic flow stability as 16.8%. Experimental results demonstrate that under CAV penetration rates of 30%, 60%, and 90%, the disturbance transfer rates of heterogeneous traffic flow are reduced to 0.85, 0.70, and 0.52, respectively, with corresponding stability margins reaching 0.15, 0.30, and 0.48. The proposed control strategy can effectively suppress disturbance propagation under relatively low penetration rates, providing a feasible solution for intelligent management and control of highway bottleneck sections.