Decision-Focused Learning via Neural Optimization Layers for Battery-Assisted Grid-Import Smoothing in AI Data Centers
Read the original on arXiv Machine Learning →The paper introduces Decision‑Focused Learning via Neural Optimization Layers (BS‑DFL) for battery‑assisted grid‑import smoothing in AI data centers. It formulates a risk‑aware battery energy storage system scheduling problem that balances reserve commitments, demand‑responsive dispatch, operating costs, grid‑ramp penalties, and conditional value‑at‑risk. By training a constraint‑aware neural optimization layer directly on the scheduling objective, the method eliminates the need for optimal‑solution labels and enables fast, solver‑free rolling‑horizon scheduling, achieving a 71.70% reduction in minute‑to‑minute grid‑import changes and an 18.01% cost savings in a 10,000‑minute simulation.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.