Cooling plant case

Pharmaceutical cooling plant

AI operating-period average plant efficiency improved from 0.74 to 0.65 kW/ton, with BAS-compatible control bands, equipment sizes, daily operating data, frequency behavior, and calculation detail shown below.

Plant efficiency improvement: 12.11%
AI operating window: 37 days
Average plant efficiency: 0.74 -> 0.65 kW/ton

Plant inventory

Equipment brands, capacities, and active plant topology

The site uses a secondary-pump variable-flow chilled-water system. The central plant includes four chillers with one standby unit, four chilled-water pumps, four condenser-water pumps, three air-side chilled-water pumps, and four process-side chilled-water pumps. Normal operation is one chiller paired with one pump.

Equipment	Brand	Specification	Status
Centrifugal chiller	Trane	Cooling capacity 650RT / Power 402kW	Running
Chilled-water pump	Grundfos	Head 20m / Power 30kW / Flow 350m³/h	Running
Condenser-water pump	Xylem	Head 22m / Power 45kW / Flow 190m³/h	Running
Cooling tower	BAC	Flow 500m³/h / Power 22kW	Running

Control and BAS compatibility

The AI layer stayed compatible with existing group control

The platform supports one-click switching between existing group control and AI control. Operators can also apply AI recommendations to group-control setpoints for condenser-water pump frequency and cooling-tower frequency.

Condenser-water pump band: 42Hz to 50Hz, constrained by low-flow protection.
Cooling-tower fan band: 22Hz to 50Hz, optimized against plant energy and indoor-temperature response.
Current optimization scope: Condenser-water pumps and cooling towers.
Reserved future interfaces: Chilled-water supply temperature and chilled-water pump frequency.
Data source: BAS operating records, daily system kW/ton, outdoor dry-bulb temperature, and outdoor wet-bulb temperature.

Control logic

Cooling-water pump and cooling-tower optimization loop

The control loop uses reinforcement learning: read operating state, check load and action limits, update strategy, and execute pump/tower actions.

The reward combines overall operating energy and indoor-temperature change.
The strategy directly targets condenser-side energy instead of controlling only an intermediate temperature-difference variable.

01
Optimization cycle starts
02
Read state parameters
03
Load above start threshold?
04
State parameters inside allowed action range?
05
Discretize state parameters
06
Update Q(s,a) with reward feedback
07
Use epsilon-greedy strategy for pump/tower command
08
Execute action
09
Optimization cycle ends

Jul 2 baseline condenser-water pump frequency

Fixed-frequency operation around 48Hz, 47.6Hz, and 45.4Hz.

CwPump #1CwPump #2CwPump #4

Operation stayed near fixed frequency rather than actively searching for a lower-energy point.

Sep 28 AI condenser-water pump frequency

Two running pumps moved within the 42Hz-50Hz control band.

CwPump #1CwPump #3

The AI period includes deliberate drops near mid-day and late evening, then returns to higher frequency when conditions require it.

Jul 2 baseline cooling-tower fan frequency

Tower fans held near fixed frequency levels.

Tower #1Tower #2Tower #3Tower #4

Baseline operation is nearly flat, with one visible transient dip.

Sep 28 AI cooling-tower fan frequency

Tower fans stepped down and up across the day.

Tower #4Tower #1Tower #3

AI tower operation reduced tower running energy by about 30% during the comparison window.

Calculation detail

Plant efficiency improvement and operating-period calculation

Baseline period: Jun 1 to Jul 7 baseline operation; 27 valid daily records after missing-power-data exclusions.
Baseline averages: 0.74 kW/ton, outdoor dry-bulb 22.66C, outdoor wet-bulb 19.36C.
AI period: Sep 1 to Oct 7 AI control; 37 daily records.
AI averages: 0.65 kW/ton, outdoor dry-bulb 25.64C, outdoor wet-bulb 22.43C.
AI-period highest plant efficiency: 0.73 kW/ton, from the daily operating data.

1. Baseline average plant efficiency = 0.74 kW/ton.
2. AI-period average plant efficiency = 0.65 kW/ton.
3. Reduction = 0.74 - 0.65 = 0.09 kW/ton.
4. Plant efficiency improvement = (0.74 - 0.65) / 0.74 * 100% = 12.11%.

Case result

Plant efficiency improvement

=Baseline average kW/ton-AI average kW/tonBaseline average kW/ton*100%

=0.74-0.650.74*100%

=12.11%

Efficiency benchmark

U.S. reference context for plant efficiency

U.S. chilled-water plant practice usually compares measured plant performance in kW/ton. The table gives ASHRAE reference context without treating it as a certification claim.

Metric	Baseline	AI period	U.S. reference context
Plant efficiency	0.74	0.65	Lower is better; this is the main U.S. operating-efficiency view for chilled-water plants.
ASHRAE reference context	0.61-0.70 kW/ton	0.61-0.70 kW/ton	ASHRAE District Cooling Guide lists electric centrifugal chillers in this typical range.
Measurement boundary	System kW/ton	System kW/ton	Used as reference context, not a U.S. certification score or whole-building energy benchmark.

Daily operating data

Daily plant efficiency, dry-bulb, and wet-bulb records used for the scatter plots

The table below lists daily operating records, and the two plant-efficiency weather scatter plots use kW/ton values from these points.

Plant efficiency vs wet-bulb temperature

Baseline operationAI control

Plant efficiency vs outdoor dry-bulb temperature

Baseline operationAI control

Mode	Date	kW/ton	Dry-bulb C	Wet-bulb C
Baseline operation	Jun 1	0.71	21.27	18.27
Baseline operation	Jun 2	0.73	22.46	17.78
Baseline operation	Jun 3	0.67	20.36	15.11
Baseline operation	Jun 8	0.69	21.29	17.67
Baseline operation	Jun 9	0.70	23.23	18.43
Baseline operation	Jun 10	0.71	23.06	17.79
Baseline operation	Jun 11	0.69	21.24	17.54
Baseline operation	Jun 14	0.69	20.80	17.49
Baseline operation	Jun 15	0.69	21.57	16.96
Baseline operation	Jun 16	0.68	20.71	16.87
Baseline operation	Jun 17	0.69	18.25	17.24
Baseline operation	Jun 18	0.73	19.30	18.77
Baseline operation	Jun 19	0.76	20.52	19.80
Baseline operation	Jun 20	0.77	22.31	19.73
Baseline operation	Jun 21	0.75	23.58	18.96
Baseline operation	Jun 22	0.72	22.69	17.77
Baseline operation	Jun 23	0.70	20.24	17.18
Baseline operation	Jun 24	0.71	18.11	17.63
Baseline operation	Jun 25	0.76	21.17	19.72
Baseline operation	Jun 26	0.76	21.62	19.60
Baseline operation	Jul 1	0.79	23.86	20.88
Baseline operation	Jul 2	0.82	24.27	21.66
Baseline operation	Jul 3	0.82	24.67	21.84
Baseline operation	Jul 4	0.82	26.12	21.82
Baseline operation	Jul 5	0.81	25.11	21.78
Baseline operation	Jul 6	0.85	31.39	27.41
Baseline operation	Jul 7	0.85	32.54	26.95
AI control	Sep 1	0.64	26.10	21.77
AI control	Sep 2	0.63	27.06	22.70
AI control	Sep 3	0.64	26.96	23.74
AI control	Sep 4	0.67	28.69	25.27
AI control	Sep 5	0.70	28.18	25.15
AI control	Sep 6	0.67	28.71	24.63
AI control	Sep 7	0.64	28.15	24.29
AI control	Sep 8	0.63	28.17	23.43
AI control	Sep 9	0.63	27.57	22.85
AI control	Sep 10	0.62	27.78	22.56
AI control	Sep 11	0.64	28.80	24.47
AI control	Sep 12	0.67	27.48	24.90
AI control	Sep 13	0.66	25.50	23.95
AI control	Sep 14	0.63	24.65	23.29
AI control	Sep 15	0.63	24.42	23.51
AI control	Sep 16	0.65	25.30	23.86
AI control	Sep 17	0.65	27.73	23.87
AI control	Sep 18	0.69	29.24	25.74
AI control	Sep 19	0.72	30.19	26.60
AI control	Sep 20	0.69	26.86	24.41
AI control	Sep 21	0.63	23.20	20.68
AI control	Sep 22	0.61	21.04	19.38
AI control	Sep 23	0.62	20.42	18.89
AI control	Sep 24	0.64	23.11	22.05
AI control	Sep 25	0.65	26.63	23.95
AI control	Sep 26	0.66	25.63	23.42
AI control	Sep 27	0.72	27.62	24.71
AI control	Sep 28	0.72	25.80	24.58
AI control	Sep 29	0.73	27.40	24.55
AI control	Sep 30	0.71	24.15	22.09
AI control	Oct 1	0.67	23.62	19.32
AI control	Oct 2	0.59	22.95	17.60
AI control	Oct 3	0.62	22.63	18.51
AI control	Oct 4	0.66	23.54	18.88
AI control	Oct 5	0.61	22.26	16.79
AI control	Oct 6	0.60	21.45	16.93
AI control	Oct 7	0.61	19.55	16.60

Run the same measurement discipline on your cooling plant

We can review BAS points, operating history, equipment sizes, and metering coverage to determine whether a similar AI supervisory layer is practical.

Book a demo Back to cases