1. Central chiller plant
Instead of every building running its own rooftop units, one central plant houses large chillers. Consolidating production allows higher-efficiency machinery, better maintenance, and load diversity across many buildings.
TECHNICAL GUIDE
How AI-powered district cooling works: a guide for cognitive cities
District cooling produces chilled water centrally and pipes it to many buildings instead of each one running its own equipment. This guide explains the physical chain end to end, the numbers operators actually watch, and where forecasting changes the job from reacting to conditions to anticipating them.
THE PHYSICAL CHAIN
From central plant to occupied space.
2. Chilled-water distribution
The plant chills water — typically to around 4–6 °C — and pumps it through insulated underground supply pipes to connected buildings. Warmer return water flows back to the plant to be chilled again.
3. Energy transfer station
At each building, a heat exchanger transfers cooling from the district loop to the building's internal loop. This keeps the two circuits hydraulically separate and gives a clean metering point for billing.
4. Building distribution
Air-handling units and fan coil units use the chilled water to cool air delivered to occupied spaces. Valves modulate flow so each zone draws only the cooling it needs.
5. Heat rejection
The heat removed from buildings is rejected through cooling towers or condenser water systems. In hot, humid Gulf conditions this stage strongly influences overall plant efficiency and water consumption.
Why cities choose it
Centralising production removes equipment from individual rooftops, concentrates specialist maintenance in one place, and lets diversity between buildings reduce the total capacity a district has to install and power.
THE NUMBERS THAT MATTER
Four measures operators live by.
- kW/TR
- Kilowatts of electrical input per ton of refrigeration delivered. The core efficiency measure for a cooling plant — lower is better, and it varies with ambient conditions and part-load operation.
- Delta-T
- The temperature difference between chilled-water supply and return. Low delta-T means the loop is moving a lot of water for little cooling, which wastes pumping energy and eats into capacity.
- Ton of refrigeration (TR)
- The standard unit of cooling capacity used across district cooling contracts, roughly 3.517 kW of heat removal.
- Capacity utilisation
- Average delivered load against installed capacity. Persistently low utilisation signals stranded capital; persistently high utilisation signals limited resilience headroom.
REACTIVE VS PREDICTIVE
Where forecasting changes the operating model.
Traditional district cooling management is largely reactive: the plant responds once conditions have already moved. A predictive layer uses weather forecasts and observed plant behaviour to shift decisions earlier in the day.
| Traditional reactive management | Predictive, forecast-led operation |
|---|---|
| Operators respond after temperatures drift or an alarm fires. | Forecast load and ambient conditions hours ahead, so staging changes before drift occurs. |
| Chillers staged on fixed rules or operator habit. | Staging informed by expected demand curve, equipment efficiency profile, and weather. |
| Low delta-T noticed during periodic reviews. | Continuous detection of degrading delta-T with the affected branch or building flagged. |
| Maintenance scheduled by calendar interval. | Attention prioritised by observed drift in efficiency and behaviour of individual assets. |
| Resilience assessed after a heat event or outage. | Headroom against forecast peak reviewed continuously as a planning input. |
CoolGrid Dubai is at development and pilot-exploration stage. This guide describes how the approach works; it does not claim verified savings for any specific site.
WHAT THE MODELS READ
Six signals behind a cooling forecast.
Weather feeds
Ambient temperature, humidity and forecast horizon — the strongest external driver of cooling demand.
BMS telemetry
Zone temperatures, valve positions, air-handling unit behaviour and setpoints from building management systems.
Plant instrumentation
Chilled-water supply and return temperatures, flow, and chiller status from the central plant.
Energy and water meters
Electricity and make-up water consumption, used to derive intensity and efficiency measures.
Maintenance records
Service history and interventions, giving context to changes in equipment behaviour.
Occupancy signals
Schedules and building usage patterns that shape when and where demand appears.
See the forecasting layer running on live Dubai weather.
The live demo pulls real hourly Dubai weather and shows demand, delta-T and efficiency behaviour across clearly labelled simulated cooling sites, with an AI operations briefing on top.