Forecast the peak. Shave it before it forms.
GEST reads the campus load curve with an LSTM model, predicts demand two hours ahead, and trims 20 kW off the peak automatically.
Loading scene…
Aging plant, volatile load, no foresight
A lecture hall peaks at 120 kW; a dormitory holds 80 kW at the base. Forty percent of HVAC on campus predates 1990, and a quarter to a third of the energy supplied is lost to systems that only react after the fact.
Lecture hall peak
Dormitory base load
HVAC predating 1990
Energy lost to waste
Five subsystems, one controller
LSTM forecasting
95% accuracy · 2 kW MAE · 20 kW peak shave (8am to 2pm)
The model learns each building on its own terms: the morning surge in the lecture halls, the flat overnight draw in the dormitories, the unscheduled spikes from lab equipment. It calls the next move before the campus makes it.
IoT sensing mesh
300+ sensors · 1 Hz · MQTT
Once CO₂ passes 600 ppm, the controller pulls 5 kW from lighting and 8 kW from HVAC on its own. Telemetry streams continuously over MQTT rather than an hourly poll, so the model never reacts to a campus that has already moved on.
Phase-change thermal storage
SiC-doped · 0.8 W/m·K · 50 kJ/in³ · COP 3.0 to 3.3
Silicon-carbide-doped storage banks absorb daytime heat and release it on demand. That alone lifts chiller COP from 3.0 to 3.3 and moves cooling load off the peak.
Piezoelectric corridors
Pb(Zr,Ti)O₃ · 0.1 MPa/step · 5 W/m²
PZT tile underfoot converts footfall into 5 W/m², enough to run the sensor nodes watching that same corridor. The hallway powers its own instrumentation.
Perovskite windows
1.5 eV · 15% efficiency · 70% transmission · 5 W/m²
A semi-transparent perovskite glaze generates 5 W/m² while still passing 70% of the daylight through. The building envelope produces power without dimming the room behind it.
Drive the control layer yourself
This is the building-management view a facilities director would sit in front of, running on simulated telemetry. Trip the load-shed controls and watch the forecast peak flatten in response.
Loading instrument view…
What one year of GEST returns
Energy loss reduction
Saved per year
Annual cost saved
CO₂ avoided per year
Demand charge cut
per month
Grid voltage at peak
Built for these operators
- 01University campuses running pre-1990 HVAC against a 2030 efficiency mandate
- 02Research facilities where lab equipment spikes the load without warning
- 03District energy operators keeping demand charges predictable
Questions facility teams ask
01How far ahead does GEST forecast electrical demand?
Two hours. An LSTM model reads more than 300 sensors at 1 Hz over MQTT and predicts campus demand at 95% accuracy with a mean absolute error of 2 kW, measured in our pilot deployment.
02How much load can GEST actually shed?
Up to 20 kW across the 8am to 2pm peak window. When CO₂ in occupied zones passes 600 ppm, the controller dims non-critical lighting for 5 kW and raises the cooling setpoint for another 8 kW, automatically and ahead of the peak.
03What hardware does GEST rely on?
An IoT sensing mesh, silicon-carbide-doped phase-change thermal storage that lifts chiller COP from 3.0 to 3.3, piezoelectric corridor tile generating 5 W/m² from footfall, and semi-transparent perovskite window glazing at 15% efficiency.
04Is the dashboard on this page showing real data?
No. The control-room view runs on simulated, illustrative telemetry so you can explore the interface. The performance figures on this page come from pilot measurement, and we share the underlying data on request by email.
Want the forecast accuracy on your own load curve?
Send us a week of campus metering and we will show you where the peak forms, and how much of it GEST removes.