Objective: Urban expressway interchanges with closely spaced diverging and merging areas are typical recurrent bottlenecks under high demand. Mandatory divergence toward off-ramps and merging from on-ramps induce intensive weaving conflicts, and local disturbances may propagate across adjacent bottlenecks, resulting in system-level congestion. Conventional traffic management approaches (e.g., ramp metering and variable speed limits) are usually designed at a macroscopic level and often target isolated bottleneck segments; therefore, they can mitigate local congestion but may not effectively suppress congestion coupling between successive diverge–merge areas within an interchange network. With the development of intelligent connected vehicles (ICVs) and cloud control systems, vehicle-level cooperative decision-making has become feasible. Nevertheless, existing studies predominantly address either a single merging zone or a single diverging zone, while integrated interchange-level coordination of diverging and merging operations remains insufficient. To bridge this gap, this paper proposes an interchange-level cloud-based cooperative optimization framework that jointly coordinates diverging and merging operations. The objective of this study is to compute a near-optimal passing sequence and trajectory plan that minimizes the weighted sum of travel delay (WD) for all vehicles while ensuring safety and comfort. The main challenge lies in the combinatorial complexity caused by multilane interactions and mixed mandatory and discretionary lane changes. Methods: To improve computational scalability, a rolling traversal scheduling (RTS) mechanism is developed. Instead of solving a single large-scale optimization for all vehicles at the interchange, the RTS constructs a rolling decision group comprising the current leading vehicle (or platoon) from each relevant lane. At each decision step, the cloud controller formulates a mixed-integer linear programming (MILP) subproblem to determine which candidate to serve next in the conflict area, along with the corresponding discrete sequence decisions. Once the decision is fixed, the selected candidate leaves the group, the next vehicle in that lane enters, and another MILP is solved. Through this rolling update, the interchange-level scheduling task is decomposed into a sequence of tractable MILP subproblems, while the cost design accounts for the delay impacts on other vehicles to preserve near-global optimality. To further enhance multilane utilization, a free lane-change strategy for through vehicles is proposed that jointly searches for discretionary lane-change positions alongside mandatory lane-change requirements to create gaps and reduce conflicts. In addition, a dual-mode trajectory planning method is introduced to translate the optimized sequence into implementable motion: an optimal-control-based trajectory is generated for efficiency and smoothness, while a car-following-model-based trajectory is retained as a safety-guaranteed fallback. Results: A simulation platform is implemented on an interchange segment with stochastic arrivals under 720, 1 080, and 1 440 pla·h^-1·ln^-1. Compared with a baseline strategy that optimizes diverging and merging areas separately, the proposed method reduces WD by 22.5% in an illustrative case, exhibits considerably better resilience at 1 080 pla·h^-1·ln^-1 (delay increase 33.7% versus 320.0%), and maintains a relatively fluent state at 1 440 pla·h^-1·ln^-1 with a 24.6% higher average speed and markedly lower average delay (~15.5 s versus 55.0 s). Conclusions: Overall, the proposed framework integrates rolling MILP scheduling, discretionary lane-change coordination, and robust trajectory planning, demonstrating the potential of cloud-controlled ICV coordination to mitigate congestion coupling and enhance interchange operational efficiency.