What we’re actively working on.

Sub-feeder load attribution under DER ambiguity

Behind-the-meter solar, batteries, and EVs make it hard to disentangle real load from gross load on a feeder. We're benchmarking iTransformer + N-BEATS-X variants against a behind-the-meter Bayesian decomposition.

Grid-forming control under low-inertia regimes

IEEE 2800-2022 negative-sequence requirements for GFM IBRs are non-trivial under asymmetric LVRT. We're prototyping reinforcement-learning controllers that satisfy the standard while minimising overcurrent.

Acoustic provenance for transformer monitoring

PANNs embeddings work zero-shot, but each transformer has its own acoustic fingerprint. We're studying few-shot enrolment protocols that reach 95% partial-discharge precision with under 30 minutes of audio per asset.

VPP bidding under privacy constraints

FERC 2222 lets DER aggregations clear in capacity markets, but customers won't surrender raw interval data. We're working with secure-aggregation primitives so the aggregator only ever sees masked sums.

Weather-driven outage forecasting

GraphCast skill at 6h is excellent for storm trajectory but coarse for ignition risk. We're fine-tuning on PG&E-style PSPS labels to sharpen the link between forecast skill and operational decisions.

5GDHC scheduling with thermal storage

Bidirectional district networks make scheduling a coupled hydraulic-electric problem. We're applying Temporal Fusion Transformers + MILP to compress 24-hour schedules to seconds of solve time.