Forecasting that
shows its work.
Adaptive load forecasting for rural electric cooperatives — built around a pre-registered, falsifiable evaluation protocol. We publish the receipts: regime detections, calibration coverage, baseline comparisons, and the failure modes we're working on. No leaderboard hacking, no buried numbers.
Real-Time Grid Demand
PJM Interconnection · Last 7 days · EIA API
Data: U.S. Energy Information Administration Open Data API
Full research data →Live Platform Statistics
18,724
Hourly Load Readings
14,905
Forecasts Generated
1,056
Forecasts Graded
5
Regime change-points (last 209d)
Live from the LoadLens database. Regime count is from php artisan loadlens:find-regime; eval metrics, calibration coverage, and per-regime MAPE breakdown live on the methodology page.
How It Works
Four components. One closed loop.
Ingest
Real-time hourly demand data from EIA, NOAA weather, and cooperative smart meters.
Predict
Multiple independent forecasting models run simultaneously — each specializing in a different signal type.
Grade
Every prediction is automatically scored against actual outcomes. Per-model accuracy tracked continuously.
Adapt
Model weights are recalculated based on recent performance. The system learns which models work best in current conditions.
Distributed solar reshapes the load curve. LoadLens's CUSUM detector flags the regime change as it happens — not weeks later in a quarterly review.
What Makes LoadLens Different
Pre-Registered Eval
Every accuracy claim on this site comes from php artisan loadlens:eval against a versioned, pre-registered protocol. Holdouts, baselines, regime stratifications, and falsification conditions all declared in advance — not retrofitted after the fact.
Calibrated Probabilistic Forecasts
Point forecasts hide uncertainty. LoadLens ships [q10, q90] intervals via split-conformal calibration — the kind of bounded reserve-margin signal coop dispatchers actually run procurement on. We publish the calibration error openly.
Cooperative-Priced
Enterprise analytics platforms cost $200K–$2M/year and target utilities with 500K+ meters. LoadLens is built for cooperatives serving 5,000–50,000 members at a price they can actually justify.
Regime Detection · Real Receipts
5 statistically significant change-points in the last 209 days.
Sweep of PJM Region hourly load with a CUSUM threshold of 8.9. Top-3 detections (by CUSUM magnitude) below — each row is a real moment when the load distribution shifted enough to flag a regime change, with before/after MAPE for the live ensemble computed on 48-hour windows on either side.
| When | CUSUM | Δ load | Kind |
|---|---|---|---|
| 2026-05-20 02:00 | 178.4 | +32.9% | peak_demand |
| 2026-02-12 07:00 | 95.7 | -13.6% | reduced |
| 2026-03-19 10:00 | 90.9 | +17.3% | elevated |
A single Tesla at home adds an entire house's average load. EV adoption arrives in clustered bursts that move the statistical regime. The CUSUM detector picks them up; the methodology page logs every detection openly.
Built For
832
Rural electric distribution cooperatives in the U.S.
42M
Americans served across 56% of U.S. land area.
1
Affordable adaptive forecasting option for coops under 50K meters.
Ready to put your cooperative's data through the same eval?
Upload your smart meter CSV and we'll run the same pre-registered eval against your distribution-level load — including the regime stratification, calibration check, and a copy of the receipt for your engineering team. Free 90-day pilot for qualifying cooperatives.
Built by Champlin Enterprises LLC · (815) 885-5509 · contact@champlinenterprises.com