Generative Smart Energy Management in Buildings Using IoT Sensors: An Optimization-Based Approach

Abstract

Artificial Commercial and institutional buildings use a significant amount of electricity, much of which is wasted because heating, ventilating and air-conditioning (HVAC) and lighting systems are set on fixed schedules instead of according to need. The study aims to explore the effectiveness of IoT-based energy management in buildings based on experiments and optimization methods. An electricity sensor network was installed in a simulated three-zone building model (1200 m², offices, meeting rooms and a laboratory/classroom space) and measured the electricity consumption, indoor temperature, humidity, occupancy, status of HVAC systems, illuminance and other sensor readings every 15 minutes over a period of 28 days. The observations were summarized in time series, abnormal and incomplete readings were filtered and analyzed to describe the consumption patterns. A rolling-horizon linear optimization model was then formulated to determine the cooling power, fan speed, lighting dimming levels under the constraints of comfort, ventilation, and capacity with a penalty for peak demand. The five different operation scenarios under consideration were compared with a baseline based on a schedule: Occupancy rules, Optimized HVAC, Optimized lighting and Integrated optimization. The integrated approach resulted in an overall reduction in electricity consumption of 26.4%, and a reduction of 35.4% in controllable-load electricity consumption, a 36.4% reduction in peak demand, and a 98.5% compliance with the 22–26 °C comfort band as measured by the occupant weighted time index. There was a statistically significant amount of savings on each day (paired t(27) = 6.57, p < .001). A sensitivity analysis indicated that there was a definite relationship between the energy savings and reduction of peak. The results are an upper bound, as the study was performed with simulated sensors and an optimizer with a complete knowledge of the building model, and will be validated during a physical deployment.