Early warning for grid anomalies, days before the event.

What it does

HTM-Monitor watches a grid's own telemetry — demand, net generation, forecast error, imbalance — and flags when the system is behaving in ways it hasn't behaved before. No fixed thresholds. No supervised training. The model learns each operating signature as it runs and raises a signal when real-time data departs from the learned envelope.

The goal is simple: surface grid stress early enough to matter, before fixed thresholds tuned for last season's fuel mix catch the event retrospectively.

Validation — targeted reliability events

*Onset detection rather than forecast — flagged once the event was underway. Full methodology, per-ISO results, and complete alert logs available on request.

How it works

The engine is Hierarchical Temporal Memory — a biologically-inspired sequence-learning algorithm — extended with a grouped-consensus decision layer that only raises a system-level alert when multiple independent signals agree. Unsupervised, online, and auditable per step.

Runs single-process in Python 3.9+. No GPU. No external calls. Deploys on a $5/mo VPS or inside an air-gapped VM — operator's choice.

Ways to start

Technical preview — free. Send a 90-day telemetry sample and one past event you'd like scored. You get a preview audit inside a week: lead/lag, false-positive rate, per-signal notes. No commitment.

Paid pilot — $2,500. Full historical replay against your events, with a written per-event audit inside a week. The fee credits in full against your first month of subscription.