Machine learning can address the design of building services in new buildings by optimising the efficiency of different energy management.


Using machine learning to ‘green’ our buildings

We live in a world of buildings, in our cities, in our towns. As our global population grows, to almost ten billion by 2050, even more buildings will be needed. Although incredibly important to our day-to-day lives, they come with an environmental impact.

Buildings consume nearly 40 per cent of end-use energy worldwide and are responsible for approximately one-third of total carbon emissions. Operational emissions (from energy used to heat, cool and light buildings) account for approximately 28 per cent of global emissions.

Sustainability – an environmental, economic, social and governance imperative

Sustainability has become a focus for governments and investors globally and if you are a building owner and/or operator you should be considering how to improve the sustainability of your assets from a multi-dimensional perspective:

  • Environmental – Do you want to reduce your carbon emissions?
  • Economic – Do you want to decrease your energy costs through a simple, low risk, zero cost way to realise immediate operational savings?
  • Social – Do you want to improve your carbon footprint and the impact you are making on your community and the world as a whole?
  • Governance – Are you focused on leadership in your sector and maintaining or gaining competitor advantage and market share through innovation?

Optimising building systems to increase efficiency and decrease carbon emissions

In this thinking paper, we consider a bold vision for how building owners and operators can create an enviable sustainability brand and establish market leadership by reducing operational carbon emissions and energy costs, noting that 75 per cent of a building’s lifecycle costs are in the operational phase. It’s using machine learning to optimise building systems by considering existing data to predict future scenarios and uncover opportunities for improvement.

BOTF2 lifecycle graph

BOTF2 lifecycle graph mobile

While the traditional rules of thumb and building standards may improve building energy efficiency, they don’t necessarily result in optimal operation as they don’t consider the interactions among multiple pieces of equipment. Machine learning and artificial intelligence algorithms can forecast and improve building energy performance.

The machine-learning application considers energy efficiency of existing stock with existing data to demonstrate how the system has historically performed over time. A move to decrease a building’s operational emissions in existing stock requires careful monitoring of energy consumption and building performance.

BOTF2 lifecycle graph

The first step in optimising building energy consumption is calculating the amount of carbon emissions by using a building energy assessment method. Figure 1 shows how the assessment is separated into four parts. This is an informative method that provides a comparative energy performance index for decision-makers.

Building energy assessment method

Building energy assessment method

Improving energy prediction and analysis through machine learning

Machine learning is generally used to describe a computer algorithm that learns from existing data. Machine learning models discover the relationship between multiple components with regards to a target variable. When the algorithm is trained on enough data, of high enough quality, it can make accurate predictions on the energy consumption and performance of different system control methods.

The multidimensional interactions between system components make this process impractically difficult via traditional analytical means.

There are opportunities for building owners and developers to improve energy prediction and analysis using machine learning. The outputs enable much higher resolution decision-making around investing in building services systems that lower carbon emissions and energy costs. It is the precise application of complex mathematics and building dynamics.

Historical and current building energy data is used to derive accurate energy usage for all building components with internal and external details as the inputs (e.g. climate information, construction fabric, heating and cooling, water).

Applying machine learning to existing and new buildings

Machine learning can address the design of building services in new buildings by optimising the efficiency of different energy management, heating, ventilation, air conditioning and water systems. It may provide a more detailed alternative to traditional building energy benchmarking and rating schemes.

For existing stock, machine learning uses a data-driven model trained on historical performance data. In this way, building data is used to define possible energy parameters of new materials, equipment and systems (figure 2).

Machine learning pathway

Machine learning pathway

Accurate estimation of heating and cooling load is the foundation of successful design for any HVAC system, and this leads to reduced operational costs through saving energy consumption. Advanced forecasting of electricity loads allows determination of excessive usage periods, reduced peak demand and the load profile of an electrical HVAC system.

Machine learning algorithms can optimise chilled water systems to reduce their energy consumption. A chilled water system allows for desired cooling requirements to be maintained inside a building and each system is created differently to suit the individual building.

A typical commercial office building system can use up to one million kilowatt hours of electricity annually, which is approximately 40 per cent of a building’s total energy use. Air conditioning is therefore a major contributor to carbon emissions.

Case study: optimising chiller plant in a large commercial building

Aurecon recently studied a commercial building’s large chiller plant system in South East Queensland containing five chillers of varying age and performance. Machine learning was applied to determine its possible energy usage and energy saving potential. We used existing measured component data, collected at six-minute intervals over the duration of 13 months, to inform calculations.

A deep, multi-layered, perceptron model architecture was developed to accurately calculate the possible electrical energy consumption of each chiller and its associated pumps in the system (figure 3).

Methodology for optimising chiller energy consumption through machine learning

Methodology for optimising chiller energy consumption through machine learning

Our process calculated the following possible savings by changing the control logic of how the system stages the total cooling demand (figure 4). In total, an 800-tonne reduction in carbon emissions could be achieved annually.

Changing the control logic

Changing the control logic

This analysis is a big step forward for deep learning in building services system design and optimisation to respond to climate change challenges.

Low risk, zero cost option machine learning model

A low risk, zero cost option to assess and achieve potential benefits from a machine learning approach can involve a ‘share of savings’ commercial model. Aurecon has developed an approach that draws on our engineering skills, domain experience and data science expertise, together with a commercial model to:

  1. Asses existing plant
  2. Collect data
  3. Confirm the commercial model – percentage of savings and term defined. If no benefit found, no cost to building owner/operator
  4. If viable to proceed – undertake machine learning phase
  5. Deploy – including quarterly reviews
  6. At end of term – savings realised for building owner/operator

Towards net zero carbon with machine learning

Optimisation of building services systems to reduce energy usage and costs will continue to gain attention as the sector contributes significantly to environmental pollution and fossil energy consumption.

As a result, machine learning and artificial intelligence is becoming more widely used to satisfy the demand for fast and accurate forecasting, which is essential for investment decision-making. Machine learning models are required in the sector to answer this demand and optimise building services systems.

Machine learning operates closely with other digital technologies such as big data, analytics, data management, visualisation, augmented reality and virtual reality as we create smart infrastructure for the future.

The transition towards mainstream net zero carbon standards requires immediate action to achieve greater innovation and improved processes to calculate, track and report operational carbon emissions.

We will see machine learning and artificial intelligence algorithms become more commonly used to explore maximum potential energy savings, identify the optimal values of design variables and provide building designers with more efficient methods of designing robust energy-optimised buildings.

Ultimately, machine learning can provide building owners and operators with a low risk, no cost opportunity to satisfy environmental, economic, social and governance imperatives and gain a market-leading position through innovation.

About the Author

Luke Mckenzie is a mechanical engineer specialising in ecologically sustainable design in the built environment.

Luke’s capabilities in environmental modelling software enable him to work closely with clients, undertaking in-depth analysis of sustainable building performance indicators including energy use, daylight levels and thermal comfort conditions. His experience spans a range of industry sectors including defence and healthcare, commercial, industrial and public buildings.

Unfortunately, you are using a web browser that Aurecon does not support.

Please change your browser to one of the options below to improve your experience.

Supported browsers:

To top