Smart Battery Swap System Optimization: Advancing Sector Coupling with Artificial Intelligence for Nearly Zero-Energy Buildings

Taiwan has the world's densest network of electric-scooter battery-swapping stations — Gogoro alone runs thousands of them. Each station holds dozens of charged batteries that mostly sit idle between swaps: in effect, distributed energy storage scattered across the city. Buildings, meanwhile, increasingly need flexibility to soak up rooftop solar and trim peak grid demand on the road to nearly zero-energy buildings (NZEB). This completed NSTC project (2024–2025) asked whether swapping stations can double as flexible energy assets for the buildings next door, coupling the transport and building sectors.

The obstacle is real-time control. Conventional optimization — model predictive control or mixed-integer linear programming (MILP) — produces accurate charge/discharge schedules but is too slow and too dependent on accurate forecasts for live operation, and operational data at any single station are sparse and uneven. The project set out to overcome both barriers at once.

Its Vehicle-to-Building (V2B) framework has three parts. A Bayesian phase-aligned synthetic time-series generator enriches sparse swapping-demand, building-load, and PV records while preserving their daily rhythm and variability. A MILP scheduler then computes cost-optimal charging and discharging under real operating constraints. Finally, a lightweight, optimization-trained neural network learns from those optimal schedules to reproduce near-optimal decisions directly from the past 24 hours of data — in milliseconds, with no online forecasting or solving. The whole pipeline was demonstrated on a Gogoro station and an adjacent building on the National Taiwan University campus.

Against an immediate-charging baseline, the optimized V2B strategy cut building electricity cost by about 9%, raised on-site solar self-consumption by 8.5%, and lifted building energy self-sufficiency by 21%, while the learned controller matched the optimizer at millisecond speed. Together these results show that battery-swapping infrastructure can act as a genuine, deployable distributed energy resource — offering a scalable, data-efficient control path for campus microgrids, public buildings, and future smart-energy districts, and scientific grounding for Taiwan's distributed-energy, smart-grid, and net-zero-city goals.