How Element AI Works

AI that doesn't just watch dashboards — it walks into the physical world: sensing the scene, learning its behavior, predicting the next half hour, and controlling equipment directly.

System Architecture

Perception → Edge AI → Autonomous Control

Three layers, one closed loop. Data is learned and decided on-site — nothing leaves your building.

① Perception Passive wireless sensors ② Edge AI Agent Learns & decides on-site ③ Autonomous Control Equipment responds directly Temperature & humidity mmWave occupancy radar Current & voltage Door / window / light Air & water flow 0.7B edge model millisecond local inference Tens of thousands of device fingerprints & energy models Trains on your site in ~10 days Data never leaves the building Predicts demand 30 min ahead Chillers / AC units Pumps (VFD frequency) Cooling towers AHU / exhaust fans Lighting & signboards LoRa / BLE BMS / API / 485 Continuous feedback — results verified against the meter, model keeps self-evolving

Element AI system architecture — sensing, edge intelligence and control in one on-site loop

Operating Loop

From First Site Visit to Self-Evolving Savings

1 Metering baseline Set the energy baseline 2 Local training Your site, ~10 days 3 Prediction 30-min lookahead 4 Best strategy Millions of combos 5 Auto control No human needed 6 Verify & evolve A/B tested & audited Every cycle makes the model smarter — savings keep growing after go-live

A closed loop from baseline to autonomous, self-improving control

Core Strategies

Six Signature Control Strategies

Not one algorithm — a portfolio of strategies the agent combines per scenario, per season, per hour.

STRATEGY 01

Edge-Side Approximation

The model continuously mirrors real conditions at the edge and re-tunes chillers, pumps and towers in real time — like a 7×24 senior engineer who never blinks, reacting to weather, crowds and holidays.

STRATEGY 02

Balanced Cycling

AI staggers start/stop order and run-hours across identical equipment: 30–50% longer equipment life, 60% fewer system faults, and 15–40% savings in transition periods.

STRATEGY 03

Staged Occupancy

When demand slightly exceeds one stage, conventional plants start the whole next stage. Element AI locks equipment at the lower stage and covers the gap with pumps and terminals — 25% saved on the same demand.

STRATEGY 04

Departure Prediction

mmWave sensing plus LSTM behavior models predict when a room will be vacated 10–30 minutes ahead at ≥95% accuracy — cooling winds down before the room is actually empty.

STRATEGY 05

Coordinated Group Control

Plant-side, distribution-side and terminal-side data merge into one operating picture. Capacity, transport and demand stay matched — 1+1>2 in both comfort and savings.

STRATEGY 06

Tariff & Virtual Power Plant

Against peak/valley tariffs, thermal mass is charged at off-peak prices and released at peaks — and the same flexibility can bid into grid demand-response programs for extra revenue.

The Algorithm Edge

One Framework That Contains the Classics

Mainstream HVAC control approaches each have strengths: HDDPC saves most in hot climates, reinforcement learning is most stable in cold weather and fault conditions, DPC needs the least hardware. Most vendors pick one.

Element AI's EPRA framework = HDDPC + DPC + RL — the agent selects and blends strategies per condition, then fuses occupancy, business data and weather that classical methods ignore. Better savings ceiling, more stable operation, lower hardware cost.

Physics-informed hierarchical control Reinforcement learning at the edge Scenario-based small models, not one bloated cloud

Why "private time-series data" wins

Image and language AI needs expensive human labeling. Element AI learns from multi-dimensional time-series data — temperatures, currents, occupancy — that is cheap to collect, private by nature, and endless in supply.

Event-fitting: virtual events map 1:1 to real events on your site
99.9% event-matching accuracy, second-level inference
Learns occupancy, SOP, tariffs, weather and equipment fingerprints
Self-evolves from every user correction and result report

See the Loop Run on Your Own Site

A free two-week A/B test proves the savings on your meter — before you spend a single baht.

Start the Conversation