Occupancy-Based Predictive AI-Driven Ventilation Control for Energy Savings in Office Buildings

Despite stricter global energy codes, performance standards, and advanced renewable technologies, the building sector must accelerate its transition to zero carbon emissions. Many studies show that new buildings, especially non-residential ones, often fail to meet projected performance levels due to poor maintenance and management of HVAC systems. The application of predictive AI models offers a cost-effective solution to enhance the efficiency and sustainability of these systems, thereby contributing to more sustainable building operations. The study aims to enhance the control of a variable air volume (VAV) system using machine learning algorithms. A novel ventilation control model, AI-VAV, is developed using a hybrid extreme learning machine (ELM) algorithm combined with simulated annealing (SA) optimisation. The model is trained on long-term monitoring data from three office buildings, enhancing robustness and avoiding the data reliability issues seen in similar models. Sensitivity analysis reveals that accurate occupancy prediction is achieved with 8500 to 10,000 measurement steps, resulting in potential additional energy savings of up to 7.5% for the ventilation system compared to traditional VAV systems, while maintaining CO2 concentrations below 1000 ppm, and up to 12.5% if CO2 concentrations are slightly above 1000 ppm for 1.5% of the time.