QCOEM: Quantum Cloud Orchestration with Evolutionary Multi-Objective Optimization

As quantum cloud platforms (QaaS) scale, they must manage heterogeneous quantum processing units (QPUs) that differ in noise profiles, qubit connectivity, and current queue lengths. Traditional cloud orchestrators typically use noise-agnostic heuristics—such as Round Robin or Greedy scheduling—which prioritize throughput but ignore the physical error rates of specific quantum devices. This paper investigates how to effectively route quantum circuits to maximize execution fidelity while maintaining system stability and minimizing task completion time.

The authors propose QCOEM, a framework that treats quantum orchestration as a Multi-Objective Optimization Problem (MOOP). The system maintains a real-time view of the cluster using heartbeat telemetry, which reports current queue states and calibrated hardware snapshots (gate durations, error rates, and coupling maps). To solve the scheduling problem, the framework employs NSGA-II and NSGA-III evolutionary algorithms to approximate a non-dominated Pareto front of possible schedules. Because the objective space is non-convex, the authors utilize an Augmented Achievement Scalarization Function (AASF) to map the Pareto set to a single, dispatchable schedule based on user-defined priorities (e.g., favoring speed, fidelity, or load balance).

QCOEM significantly outperforms traditional noise-agnostic heuristics in heterogeneous environments. By accounting for device-specific error rates and current queueing delays, the framework achieves approximately 30% higher mean execution fidelity compared to standard baselines. Furthermore, the framework demonstrates superior operational stability, achieving zero task rescheduling—a common failure point in noise-agnostic systems where tasks are dispatched to unsuitable or overloaded backends. The scheduling overhead remains bounded, making the approach suitable for practical cloud-native deployment.