Mixed-Criticality, Carbon-Aware Task Scheduling for Smart-Grid Edge-Cloud Computing: A Simulation-Based Heuristic

Abstract

Smart-grid edge-cloud networks host computational workloads of sharply different operational criticality, from millisecond-sensitive protection and control functions to delay-tolerant analytics and forecasting. Existing scheduling approaches address this heterogeneity only partially: dual-priority service differentiation protects delay-critical tasks but ignores energy cost, while carbon and price aware temporal deferral improves sustainability but ignores task criticality. This paper proposes a lightweight scheduling heuristic that unifies both principles, guaranteeing critical tasks unconditional first claim on compute capacity while deferring flexible tasks toward loweremission, lower-cost execution windows using a fast range-minimum lookahead over short-term carbon and price forecasts. The heuristic is evaluated through a discrete-event simulation driven by real UK National Grid ESO regional carbon-intensity data across eight independently selected weekdays in 2023, with a workload arrival rate calibrated from the Alibaba 2018 Cluster Trace, benchmarked against four baselines, and validated through an ablation study, a weight-sensitivity analysis, and a forecast-uncertainty experiment using the real FPN-BOA discrepancy distribution. The proposed heuristic eliminates critical-task deadline violations, reducing $\mathbf{C O}_{\mathbf{2}}$ emissions by 22.9% and energy cost by 7.1% relative to an unaware baseline, all differences statistically significant at $\boldsymbol{p}<\mathbf{1 0}^{-\mathbf{5}}$ under paired $t$-tests. A carbon-aware earliest-deadline-first baseline achieves a larger raw reduction (48.8%) with zero violations, so the proposed method's principal contribution is a structural reliability guarantee that holds by construction, independent of arrival-pattern coincidence.