Intelligent Transportation Systems (ITS) require efficient resource coordination across cloud-edge-end (CEE) architectures for dynamic traffic demands. Existing approaches face challenges in optimally allocating resources across distributed CEE systems due to non-stationary patterns and fixed parameters. We propose a theoretically-grounded meta-reinforcement learning framework. We formulate the problem as distributed resource scheduling across cloud data centers, edge servers, and end devices (traffic controllers), where computational resources must be dynamically allocated based on real-time traffic conditions. Our approach introduces Cloud-Edge-End Coordinated Meta-DQN (CEE-MDDQN), which adaptively optimizes resource allocation parameters through higher-order policy gradients across the distributed architecture. We establish theoretical guarantees with \(\mathcal {O}(\log T/\sqrt{T})\) regret bounds for our coordinated approach versus \(\mathcal {O}(T^{2/3})\) for non-adaptive methods. Through comprehensive experiments on distributed traffic control scenarios, we demonstrate that CEE-MDDQN achieves superior resource utilization efficiency and system performance compared to traditional centralized and fixed-parameter distributed approaches. This work advances the theoretical foundations of adaptive resource coordination in CEE systems and provides practical solutions for intelligent transportation infrastructures.