RACA Journal June 2026 | Page 29

FEATURE SERIES

AI

FROM FIXED LOGIC TO INTELLIGENT OPTIMISATION

By Eamonn Ryan

Artificial intelligence( AI) and machine learning are rapidly reshaping the built environment – but in cooling systems, their role is often misunderstood.

While the hype suggests fully autonomous buildings, the reality emerging from industry leaders is more nuanced

– and arguably more powerful. In a recent ASHRAE Journal podcast, host Kelly Barraza spoke with Michael Berger, head of research and development at Conserve It, about how machine learning is being applied to optimise HVAC systems, particularly chiller plants. Their discussion reveals a shift: AI is not replacing conventional control strategies, but enhancing them – especially where complexity, variability and competing system dynamics make traditional approaches insufficient.

Machine learning may be one of the defining technologies of our era, but its application in HVAC control remains“ in its infancy”( as at late 2024) with significant potential to reduce energy consumption and improve system performance, said Berger.

For decades, HVAC systems have relied on wellestablished control strategies, many codified in standards such as ASHRAE Guideline 36. These conventional approaches are robust, proven and effective across a wide range of applications.

However, as Berger explains, the question is not whether AI can replicate these strategies – but whether it can improve on them. The answer lies in complexity. In relatively stable scenarios – such as optimal start-stop routines or standard cooling tower sequencing – traditional control logic remains sufficient. But modern cooling systems rarely operate under static conditions. Instead, they are influenced by fluctuating loads, changing weather conditions, varying equipment performance, and increasingly, the integration of renewable energy systems.

It is in these dynamic, non-linear environments that AI demonstrates its value.

THE POWER OF PREDICTIVE OPTIMISATION At the heart of AI-driven cooling lies predictive modelling. Rather than relying on fixed setpoints or predefined sequences, machine learning systems continuously analyse data to determine the most efficient operating conditions in real time.

A key example discussed by Berger is the optimisation of chiller plant controls. Here, machine learning models predict equipment performance – such as chiller power consumption – based on both manufacturer data and real-time site data. These models are then calibrated to the specific characteristics of each installation, allowing for highly tailored optimisation.

This approach enables a fundamental shift from static control to adaptive optimisation. Instead of asking,“ What is the correct setpoint?”, the system asks:“ What is the most energy-efficient operating point right now?”

Perhaps the most compelling application of AI in cooling lies in its ability to manage trade-offs across interconnected systems. Cooling systems are inherently interdependent. Adjusting one component often affects others in complex ways. For example:

• Lowering cooling tower temperatures reduces chiller energy consumption – but increases fan energy.

• Running more chillers may improve efficiency at part load – or increase overall energy use.

• Shifting load between chillers and airside systems can produce conflicting energy outcomes.

Traditional control strategies struggle to balance these competing variables, particularly as conditions change. Machine learning, however, can evaluate multiple scenarios simultaneously, identifying the optimal balance between components.

As Berger highlights, there is often no single‘ correct’ answer – only a continuously shifting optimum influenced by load, weather and system configuration.

FROM COMPONENT EFFICIENCY TO SYSTEM EFFICIENCY This marks an important evolution in HVAC thinking. Historically, efficiency has been assessed at the component level: chiller COP, fan efficiency, pump performance. AI enables a broader perspective of system-level optimisation.

Rather than maximising the efficiency of individual components, machine learning seeks to minimise total system energy consumption.

This is particularly relevant in large chiller plants, where decisions such as how many chillers to run – or how to stage equipment – can have a significant impact on overall performance. In some cases, plants may include multiple chillers, pumps and cooling towers, creating thousands of possible operating combinations.

AI can evaluate these combinations in real time, selecting the most efficient configuration for current conditions. artificial intelligence

RACA Journal I June 2026 www. refrigerationandaircon. co. za 27