Real-time Model Predictive Control with Digital Twins and Edge Computing Technologies


  • Michael Berger Conserve It | Australia
  • Filippo Bernardello Conserve It | Australia
  • Craig Barry Conserve It | Australia
  • Pravesh Badjoonauth Conserve It | Australia
  • Shruthi Balaji Conserve It | Australia
  • Michael Lakhdar Conserve It | Australia



Energy, Design of Innovative HVAC systems for optimized operational performances, Digitization, Digitization in HVAC control & Health Monitoring


Buildings consume almost a quarter of Worlds Energy Consumption and hence are one of the major sources of emissions globally. In commercial buildings, HVAC is by far the most energy intensive system, accounting for close to half of the total energy consumption. For this reason every efficiency improvement in HVAC performance can significantly reduce the energy profile of the building, turning HVAC optimisation into a core requirement to deliver energy efficiency. Fundamental to optimising large energy consumers in today’s modern buildings is the use of Machine Learning in order to dramatically improve the energy efficiency of modern central cooling and heating plants. This paper will demonstrate the techniques that have been implemented to deliver advanced Real-time Model Predictive Control on Edge Computing solutions that don't require Cloud connectivity or significant computing power. Through the use of deep domain knowledge and advances in Edge Computing, it is possible to 'learn' highly accurate models of how mechanical machines operate and apply those models to predict and then solve complex optimisation problems for advanced control and improvements in energy efficiency. The authors will show how, through the collection of real-time sensor data, our platform has successfully reduced energy consumption and electrical demand in real buildings without compromising space comfort in any way at all. The capability to generate self-adjusting control algorithms in an on-premises scenario not only delivers significant outcomes but lowers overall Total Cost of Ownership for the end client. The absence of ongoing subscription fees further improves the economic model and the case for on-premises, real-time, model predictive control. Furthermore, the paper will demonstrate how the same Digital Twins used for Model Predictive Control can be used for anomaly detection algorithms or Fault Detection and Diagnosis as well as Predictive Maintenance and that this will create new service opportunities and business models for smart companies of the future whilst continuing to deliver optimal performance of mechanical systems.




How to Cite

Berger, M., Bernardello, F., Barry, C., Badjoonauth, P., Balaji, S., & Lakhdar, M. (2022). Real-time Model Predictive Control with Digital Twins and Edge Computing Technologies. CLIMA 2022 Conference.

Conference Proceedings Volume