How SiteIQ Forecasts Coincident Peaks
Our machine learning models ingest grid load, multi-source weather, building density, and Distributed Energy Resources (DER) data to identify the exact settlement hour that determines your charges.
Why we optimize for Operational F1, not MAPE.
Other forecasting companies optimize for MAPE — mean absolute percentage error. But MAPE alone doesn't tell you whether you caught the peak.
SiteIQ optimizes for Operational F1 — balancing Event Recall (did we capture every actual peak?) and Window Precision (how many of our alerts were correct?).
Six categories of data powering every forecast.
Every prediction is grounded in a converged stream of grid telemetry, multi-source weather, and customer-specific load data — refreshed continuously.
Grid Load Data
Real-time measurements and historical load from ISO (Independent System Operator) databases. Continuously ingested at 5-minute intervals.
Weather Data
Temperature, humidity, wind speed, wet-bulb globe temperature from 4+ commercial sources per zone.
Building Density Metrics
Derived from satellite imagery analysis using object recognition. Updated quarterly.
distributed energy resources (DER) Intelligence
Distributed Energy Resources (DER) — rooftop solar PV and battery storage data from interconnection databases and satellite detection.
Calendar & Economic Features
Day-of-week effects, holidays, industrial schedules, economic indicators.
Interval Meter Data
Customer-specific 15-minute load profiles from AMI (Advanced Metering Infrastructure) meters.
Two-stage intelligence.
Peak Identification
XGBoost regression for 7-day hourly load forecasts, plus a Temporal Fusion Transformer for peak classification.
HVAC Optimization
Reinforcement learning on edge controllers. Cloud trains the model, edge executes autonomously.
Closing the loop: from forecast to verified savings.
Our reported savings are measured outcomes, not modeled estimates.
- 1Collect utility invoice data
- 2Identify settlement load from the bill
- 3Match to customer interval data
- 4Confirm forecast window alignment
- 5Calculate avoided cost
- 6Report verified savings
Three outcomes, one methodology.
Coincident-peak accuracy
Event Recall and Window Precision for every forecasted settlement hour — not just headline MAPE — so the metric that drives your ICAP or capacity charge is the metric we optimize.
Verified energy savings
Utility-invoice-anchored measurement: interval-meter load reconciled against settlement data, so reported savings are avoided cost, not modeled estimate.
Cleaning outcomes
Coverage routes, contact time, and environmental conditions captured continuously between ATP tests — timestamped, GPS-verified, and exportable for audits.
From forecast to facility action.
1. Ingest. Grid load, weather, DER telemetry, building density, calendar features, and customer AMI meters stream into the platform continuously.
2. Forecast. The two-stage model produces 7-day hourly load forecasts and identifies the most likely coincident-peak settlement hour for each ISO zone.
3. Act. Edge controllers execute HVAC and thermostat setpoint changes autonomously during high-probability windows, coordinated with occupancy and comfort constraints.
4. Validate. After the settlement period, invoice-anchored reconciliation converts the operational response into verified avoided cost — the loop that separates modeled promise from measured outcome.