Reconfiguration
Models break when feeders and transmission states change.
Grid foundation model
Certifiable AI for grid decisions that have to hold up in the field.
WattsMIND Grid Agent is a Manifold-Informed Neural Dual-encoder for power grids: a zero-shot monitoring and control engine that transfers across changing networks, reasons from sparse measurements, and reports when an operator can trust the result.
Topology-invariant
Zero-shot transfer
Per-instance certification
Millisecond control
Manifold
Informed
Neural
Dual-encoder
Research
The technology behind WattsMIND Grid Agent
arXiv:2606.00716
Graph Transfer Learning via Shared Latent Geometry: Theory and Applications
The dual-encoder result: a teacher trained on simulator physics, a student deployed on sparse field measurements, with transfer bounded by Wasserstein proximity.
arXiv:2509.08672
Universal Graph Learning for Power System Reconfigurations: Transfer Across Topology Variations
The topology-invariant backbone: one trained model transfers to unseen reconfigurations without retraining on the new grid.
Operator cockpit
Less dashboard. More verified action.
Sparse telemetry
Grid decisions often need to hold under incomplete measurements.
Operational trust
AI must know when to predict, when to certify, and when to hand the case back for verification.
Architecture
Transfer, certify, control.
The platform is a research demonstration engine for studying how transfer, certification, and feasible control can support grid operations, planning studies, software prototypes, and market-aware grid analytics.
Grid inputs
DSS / topology
Telemetry
Loads & devices
WattsMIND Grid Agent engineManifold-Informed Neural Dual-encoder
Transferable grid representation with per-instance feasibility checks.
Research outputs
Power-flow solve
Certificate
Control action
Embedded API
Manifold-informed transfer
Maps grid structure and observations into a shared operating geometry.
Neural dual-encoder
Keeps a deployable pathway aligned with the physics pathway used in training.
Feasible control manifold
Selects constraint-satisfying actions in real time.
Research themes
Five application themes, one underlying grid engine.
Theme 01
Utilities & Co-ops
WattsMIND Grid Agent for zero-shot distribution power-flow solving and cyber-resilient control on SmartDSS systems the model has never seen, with 0 system-specific fine-tuning.
Theme 02Embedded Engines
A certifiable AI engine for grid software prototypes and research services that need faster analytics inside existing study workflows.
Theme 03Transmission & ISOs/RTOs
WattsMIND Grid Agent for transmission operating intelligence: fast voltage prediction, contingency screening, dispatch feasibility, market sensitivity, and operator verification across changing grid states.
Theme 04Energy Markets
Physics-aware congestion and dispatch intelligence for trading, asset optimization, and market analytics teams.
Theme 05Flexible Loads
Grid-aware coordination for data centers and other controllable loads that need operational flexibility.
Validation
Evidence that reads like infrastructure, not demo ware.
0system-specific fine-tuning on held-out DSS systems
multi-regimetransmission intelligence across changing grid states
1.65 msaverage forward-pass time per sample
near-optimalcyber-attack mitigation trajectory
Research collaboration