Mohamad Chehade
Publications

NEO-Grid: A Neural Approximation Framework for Optimization and Control in Distribution Grids

M. Chehade, H. Zhu

Proceedings of the 59th Hawaii International Conference on System Sciences (HICSS), 2026

A neural digital-twin approach for approximating distribution-grid physics while keeping optimization and control tractable.

HICSS arXiv PDF code BibTeX
Physical feeder NEO-Grid: a neural digital twin substation DER inverter volt-var rule Power-flow surrogate piecewise-linear ReLU network Learned volt-var rule ReLU form, learned parameters DEQ fixed-point solver implicit gradients no unrolling constant memory footprint
Figure. The feeder’s volt-var loop (power flow maps reactive power to voltage, a local rule maps voltage back) is learned as two ReLU networks and solved directly as the deep-equilibrium fixed point .

What this paper adds

  • Builds a neural approximation framework for distribution-grid optimization.
  • Targets tractable volt-var control through learned power-flow structure.
  • Connects data-driven models with interpretable control rules.

Cite

@inproceedings{chehade2026neogrid,
  title     = {{NEO-Grid}: A Neural Approximation Framework for Optimization and Control in Distribution Grids},
  author    = {Chehade, Mohamad and Zhu, Hao},
  booktitle = {Proceedings of the 59th Hawaii International Conference on System Sciences},
  year      = {2026},
  doi       = {10.24251/HICSS.2026.380}
}