I’ve seen various approaches using aggregated load profiles, regression models, and machine learning. But when detailed data from submetered categories (lighting, HVAC, etc.) is available, is there a method that clearly outperforms the others in accuracy and adaptability?

Any experiences with decomposition-based modeling?

I’m particularly interested because in our recent work we proposed a data-driven method using real building data:

https://doi.org/10.1016/j.enbuild.2014.07.024

I’d love to hear what others are using and recommend for this scenario.