ProFaaS: Serverless Dynamic Resource Scheduling for Electric Grid Forecasting

With the increasing integration of renewable energy, grid prediction tasks are facing heightened dynamics and hetero-geneity. Traditional cloud services with fixed resource allocation incur excessive costs due to resource over-provisioning. Serverless has strong resource elasticity, allowing computing resources to be used on demand, thereby improving efficiency. This paper proposes ProFaaS, an elastic forecasting system leveraging server-less computing. Our framework integrates a multi-model library and a profiler-driven dynamic scheduler to optimize task-specific Service Level Objectives (SLOs) and computational costs. A two-phase optimization mechanism is introduced: 1) Offline profiling of model performance under varying resource configurations, and 2) Online greedy algorithm selecting cost-optimal resources while meeting latency constraints. Experimental results show that ProFaaS reduces computational costs by up to 32% compared to traditional VM-based solutions. This work offers practical insights for adopting serverless architectures in energy systems, providing an efficient, scalable.