Iain Stewart, Affil.AIRAH
Combining BMS data and machine learning to deliver energy savings
The control of HVAC&R equipment is a central function of Building Management Systems (BMS). Controls based on return water temperature, which are still commonplace in HVAC&R systems throughout Australia, ignore a large portion of the available data. While this method is sufficient for delivering the required chilled water to AHUs, it often results in chilled plant inefficiency. This is because it neglects to consider two crucial chilled plant control parameters: the ambient wet-bulb temperature, and the instantaneous cooling load. This leads to higher energy consumption, and higher peak demand.
Machine-learning techniques have been applied to create a data-driven “digital twin” of the chilled plant room. This is a mathematical representation of a physical system, which allows for experimentation of different control strategies to determine their impact on system efficiency. By using data from the BMS, insights can be gained for the current operational efficiency, rather than as designed, which can differ significantly.
An optimisation algorithm is then applied, as all of the energy demands for each piece of equipment are known for all possible loads and ambient conditions. By seeding the model with the ambient conditions and cooling load, optimisation algorithms are able to calculate the ideal chiller load balancing and cooling tower fan speeds to deliver the required cooling using the least possible energy. This information forms the basis of a new optimised control strategy.
The benefit of this data-driven optimised control strategy is that peak plant efficiency is known for all possible conditions. This new controls approach has seen efficiency of the chilled plant equipment improve by 10-20 per cent, with similar reductions in peak demand.
Machine learning and process optimisation techniques have been used to incorporate insights from existing BMS data into new HVAC&R control logic that improves energy efficiency. A data-driven approach utilising existing BMS data ensures that control logic is optimised for the plant being investigated. This method is suitable for all equipment types and plant configurations, meaning that it can be applied in any building. This method of process improvement has seen chilled plant room energy and peak demand drop by 10-20 per cent when compared with existing control logic, significantly reducing energy costs.
Stewart is the founder and CEO of Exergenics, which helps businesses to reduce air conditioning and refrigeration costs with machine learning and optimisation. He has consulted corporates and government agencies on the risks and opportunities inherent in the technological disruptions that will occur as we make the transition to net zero emissions. Stewart has several years’ experience in energy modelling at Monash University and working closely with CSIRO, and a master’s degree in environmental engineering from the University of Melbourne.