Non-intrusive energy state monitoring for occupancy driven load management of residential units

Demand response (DR) strategies aim to balance power supply and demand by shifting or reducing power consumption through incentives or direct actuation. Developing DR strategies requires insights into occupancy patterns and energy use behaviors. Non-intrusive methods, such as NIOM and NILM, leverage smart meter data to detect occupancy and appliance usage, but their scalability is limited due to reliance on ground truth data or high-granularity measurements. This study explores a scalable, unsupervised approach for analyzing low-granularity smart meter data to infer occupant behavior. Using the Eco dataset, which includes smart meter readings and ground truth occupancy for five Swiss households across summer and winter, we downsampled the data to 15-minute intervals and applied K-Means clustering to identify energy states. Comparison with occupancy data showed that for some households, energy states aligned well with occupancy, indicating appliance use during occupied periods, while for others, the match was weaker due to intermittent usage patterns. Energy state analysis revealed household-specific variations, with increased winter energy consumption linked to higher occupancy or reduced daylight. Additionally, this approach identified e nergy usage p atterns, s uch a s high base l oad o r high-intensity appliance use, which may not be captured by traditional NIOM methods. Overall, energy state analysis offers a s calable way to i nfer o ccupant b ehavior and household flexibility, providing valuable insights for occupant-centric DR strategies in residential energy management.