ASHRAE Journal - August 2024 - 37

TECHNICAL FEATURE
FIGURE 6 Graphical user interface representation of the chiller performance
profile learned by the ML system on-site.
0.7
0.6
0.5
0.4
0.3
0.2
0.1
100
200
Load (Ton)
Hong Kong, based on the guidelines of the International
Performance Measurement and Verifi cation Protocol.8
Energy savings from 3.6% to 18.7% were found on
the chiller plants,9,10 depending on whether only
one setpoint was optimized or the full breadth of the
approach was deployed. These were considered to be
signifi cant by stakeholders, especially considering that
they were delivered without any mechanical upgrades
and only through controls strategy changes.
A key lesson learned is the importance of transparency
for the facility management team. They need to
know what the system will do next and understand
the rationale for the systems' decisions to properly
troubleshoot when issues occur within the chiller
plant. To facilitate this, the system can display key
information in a graphical user interface, such as ML
models learned, shown in Figure 6, or the trade-off in
real-time between equipment, as per Figure 7. The next
chiller to be started or stopped in the optimal sequence
of operation is also valuable information. Additionally, a
seamless and easy way to revert to a more conventional
sequence of operations and manual control is crucial,
as well as is the ability to override specifi c setpoints to
empower facility managers and ensure they are able to
carry out maintenance tasks.
Conclusion
Ultimately, there are many ways to use ML to enhance
chiller plant controls. The present approach is to use
ML to solve very specifi c challenges that best-practice
conventional controls strategies may not always be
optimal for, leveraging engineers' and designers'
knowledge to tailor the solution. It is intended to
operate in real-time, learn the part-load performance
profi le of the actual chiller plant and update over
time as equipment performance degrades through
300
400
wear-and-tear or is restored via maintenance. As a
result, it continuously updates controls strategies in
real time and has a proven track record of delivering
signifi cant energy savings from the chiller plant.
CHWT=44.06°F/CWT=84.92°F
CHWT=44.06°F/CWT=75.02°F
CHWT=44.06°F/CWT=64.94°F
References
1. Serale G., M. Fiorentini, A. Capozzoli, et al. 2018. " Model
predictive control (MPC) for enhancing building and HVAC
system energy effi ciency: problem formulation, applications
and opportunities. " Energies 11(3):631. https://doi.org/10.3390/
en11030631
2. Yudong, M. 2010. " Model Predictive Control for the Operation of
Building Cooling Systems. " Lawrence Berkeley National Laboratory.
3. ASHRAE Guideline 36-2021, High-Performance Sequences of
Operation for HVAC Systems.
4. Boyd S., L. Vandenberghe. 2004. Convex Optimization. UK:
Cambridge University Press.
5. Nocedal J., S. Wright. 2006. Numerical Optimization. Springer.
6. Wächter A., L. Biegler. 2006. " On the implementation of an
interior-point fi lter line-search algorithm for large-scale nonlinear
programming. " Mathematical Programming 106:25 - 57. https://doi.
org/10.1007/s10107-004-0559-y
7. Kronqvist J., D.E. Bernal, A. Lundell, et al. 2019. " A review
and comparison of solvers for convex MINLP. " Optimization and
Engineering. 20:397 - 455. https://doi.org/10.1007/s11081-018-9411-8
8. EVO. 2022. " International Performance Measurement and
Verifi cation Protocol-Core Concepts. " Effi ciency Valuation
Organization. https://tinyurl.com/5yknrc4f.
9. Berger M., F. Bernardello, C. Barry, et al. 2022. " Real-time
model predictive control with digital twins and edge computing
technologies. " CLIMA 2022 Conference. https://doi.org/10.34641/
clima.2022.368
10. Berger M. 2022. " Real-world application of machine
learning and optimal control simulation for chilled water plants. "
Australasian Building Simulation 2022 Conference .
FIGURE 7 Graphical user interface representation of the real-time optimization for
a chiller's CW flow.
Optimal Total Power
200
175
150
125
100
75
50
25
Pump Power
900
Chiller Power
1,000
Flow (gal/min)
Total Power
Optimum
A U G U ST 2 0 2 4 ashrae.org ASHRAE JOURNAL
37
1,100
Maximum Safe Flow
kW/Ton
Power (kW)
Minimum Safe Flow

https://www.doi.org/10.3390/en11030631 https://www.doi.org/10.3390/en11030631 https://doi.org/10.1007/s10107-004-0559-y https://doi.org/10.1007/s10107-004-0559-y https://www.doi.org/10.1007/s11081-018-9411-8 https://www.tinyurl.com/5yknrc4f https://www.doi.org/10.34641/clima.2022.368 https://www.doi.org/10.34641/clima.2022.368 http://www.ashrae.org

ASHRAE Journal - August 2024

Table of Contents for the Digital Edition of ASHRAE Journal - August 2024