Past talks‎ > ‎

Problems in data-driven energy demand management [SLIDES]

Adrian Albert, PhD Candidate, Electrical Engineering, Stanford University 

Demand-side management is seen as a cost-effective, environmentally-considerate alternative to investments in costly, polluting generation capacity to address the increasing demand for electricity. To better monitor demand, utility companies have deployed millions of “smart meters” that collect energy consumption data at fine (sub-hourly) time scales. Yet little understanding exists currently about how such information might be used to improve operational practices. 

This talk introduces novel static and dynamic models of individual residential user consumption that use smart meter data to identify high-potential users for targeting demand-side energy management programs. First, I use a static model to connect consumption variability to cost-of-service on the grid, and to argue that users may be segmented into typical cost classes. To characterize the temperature and occupancy dependence of consumption, I introduce a dynamic model that assumes smart meter observations to be generated by a succession of thermal regimes in a Hidden Markov Model, whereby each state has a different linear response to temperature. Using this model, I compute benchmarks that enable to characterize and rank users according to their flexibility for demand-response programs. As application of this model, I present an optimization framework in which the system operator may define day-ahead control schedules for each user's thermal appliance over up to a certain “effort budget” such that the aggregate load follows a desired outcome. Finally, I show that certain user characteristics (appliances, lifestyle) may be predicted using features computed from the data. This observation has implications for both better targeting of Energy Efficiency programs and for privacy on the smart grid.

Slide deck is attached below.
Joe Kwiatkowski,
Jun 7, 2014, 1:02 PM