<p>Performance parameters under different loads.</p>

Hybrid–cloud scheduling must balance cost, performance, and reliability; yet existing approaches often suffer from burdensome parameter tuning, a limited set of optimized QoS indicators, and high computational overhead. To address these issues, we propose an EMPA–ASA–based hybrid–cloud resource scheduling algorithm and make three contributions: 1) we realize state-driven adaptive scheduling and resource allocation via MDP + Q-learning, updating the policy online as system conditions evolve; 2) we introduce an M/M/c queueing model to quantitatively encode QoS constraints, thereby improving responsiveness and load adaptivity; and 3) we fuse EMPA with Adaptive Simulated Annealing (ASA), augmented by Lévy flights to strengthen global exploration and accelerate convergence. We implement a full prototype and conduct performance evaluations. The results show that EMPA–ASA outperforms baselines across multiple QoS metrics—including end-to-end delay, response time, throughput, and packet-loss rate—and reduces total cost by approximately 48% and 70% relative to GA and PSO, respectively; its advantages in QoS and cost are especially pronounced under high-load scenarios. These findings indicate a superior cost–performance trade-off, providing an efficient and reliable solution for hybrid–cloud resource scheduling.