Existing obstacle avoidance methods for autonomous vehicles primarily focus on safety, often at the expense of traffic efficiency and with susceptibility to trajectory prediction errors. To overcome these limitations, this study proposes a Dynamic Partitioned Cooperative Obstacle avoidance Method for Autonomous Vehicles (DPCOM-AV) in the context of expressway-connected traffic. A cooperative obstacle avoidance framework based on equivalent time loss is established by quantitatively grading obstacle risks and defining obstacle impact zones. Utilizing third-order Bézier curves, the method jointly addresses safety and traffic efficiency. Particle Swarm Optimization (PSO) is applied to optimize partition boundaries and trajectory control points, while Model Predictive Control (MPC) corrects predicted trajectories. Collaborative avoidance is triggered by lateral offset, longitudinal conflict, and target lane gap to optimize the use of road spatiotemporal resources. Numerical simulations demonstrate that, in dynamic obstacle scenarios, DPCOM-AV significantly enhances safety indicators compared to static scenarios: the average Time-to-Collision (TTC) increases and the average Deceleration Required to Avoid Collision (DRAC) decreases. Specifically, relative to an Autonomous Vehicle Risk-Perception-based Obstacle Avoidance Method (OMRP-AV), DPCOM-AV improves TTC by 23.3% and reduces DRAC by 15.28%. Compared to a Hierarchical Obstacle Avoidance Method based on Lane-Change Scenarios (HOMLS-AV), DPCOM-AV increases the minimum distance to obstacles by 9.66%-23.03%, reduces average jerk by 3.49%-11.58%, and significantly lowers the probability of collision within a finite time horizon. Further analysis reveals that the penetration rate of Connected and Automated Vehicles (CAVs) markedly influences the effectiveness of DPCOM-AV, with higher CAV penetration leading to more substantial cooperative obstacle avoidance performance.