by Pierre Pinson and Henrik Madsen
In the development of smarter energy systems, it is vital that we maximize flexibility of consumer demand. To this end, it will be of utmost importance to be able to predict the potential of electricity demand to respond dynamically to varied signals.
Some countries aim to be deriving almost all of their power from renewable sources in the relatively near future (Denmark’s timeframe, for instance, is 2050). With greater integration of renewable energy generation, demand flexibility will become ever more important in supporting smart energy systems. This will translate to a paradigm shift, from a system where demand drives generation to a system where renewable energy generation may influence demand patterns. In practice this requires enhancing, and taking full advantage of, the potential flexibility of all electricity consumers, including domestic households.
In contrast to large industrial consumers, for which direct bilateral agreements may be made and used on an ad-hoc basis, domestic consumption is a far greater challenge to manage owing to the large number of individual households, their distribution, the state of the art in ICT (Information and Communication Technologies), the effectiveness of economic incentives, behavioural effects, etc. A number of research and demonstration projects are investigating these factors, including the iPower project in Denmark, funded by the DSR-SPIR-program (project number: 10-095378, see link below).
Regardless of how demand flexibility is to be enhanced at the household level (electric heating, cooling, electric vehicles, etc.), identifying intelligent ways to alter demand patterns is a stochastic optimization or control problem, comprising a whole challenge in itself. This question will depend, to some extent, upon the time scales considered (and corresponding mathematical formulation), engineering considerations - for instance related to ICT capabilities, but also on philosophical aspects of design. The two main approaches currently under study are (i) direct distributed control, and (ii) the “indirect control approach” based on price signals. The latter takes advantage of the elasticity of consumers, ie the adaptation of consumption in response to varying electricity prices. Price signals are to be sent daily for optimal task assignment (bulk heating, washing machines, etc.), but also adapted in real time so as to take corrective action supporting the optimal matching of generation and consumption. In this indirect control by price setup, the stochastic optimization or control problem translates to issuing optimal price signals to be broadcast to groups of consumers whose consumption levels are to be influenced.
With this objective in mind, the core, and most crucial, aspect is to identify and be able to predict how small consumers respond to varying prices. We refer to this as the conditional dynamic consumer elasticity. It is conditional since the potential to affect the timing, and maybe even the magnitude, of the flexible part of the load is clearly a function of external conditions. If considering space heating for instance, outdoor temperature, as well as the settings of the local heat controller, will directly impact the potential demand response to prices. Similarly in the case of electric vehicles, the demand response potential will vary as a function of the time of the day when more or less electric vehicles may be plugged in and their batteries made available for demand response. In parallel this response is dynamic as most consumption patterns cannot be deferred indefinitely: batteries of electric vehicles need to be charged at some point before they are to be driven, while households need to be heated so as to keep indoor temperature at an acceptable level.
As a final point, this conditional dynamic elasticity of electricity consumers may smoothly evolve with time, owing to changes in consumption patterns, appliances and their functionalities, etc. As a consequence, one needs to employ a bottom-up approach and use empirical data for the identification of appropriate models, adaptive estimation of their parameters, and continuous monitoring of forecast quality. The quality of such forecasts will be paramount since this data will directly impact the reliability of potential demand response. An unreliable demand response would make this an inefficient solution compared with alternatives, such as using storage or expensive conventional generators, possibly even magnifying the fluctuations that we are aiming to dampen. Ideally, these predictions should be of probabilistic nature, in the form of scenarios, so as to fully describe the range of potential responses from the aggregation of household consumers to be influenced.
 O. Corradi et al: “Controlling electricity consumption by forecasting its response to varying prices”, IEEE Trans. Power Syst., May 28, 2012, http://dx.doi.org/10.1109/ TPWRS.2012.2197027
 D. Hammerstrom, et al: “Pacific NorthWest GridWise Testbed demonstration projects – Part I: Olympic Peninsula project”, Tech. Rep., Pacific NorthWest National Laboratory, PNNL-17167, October, 2007
 J. Torriti, M.G. Hassan, M. Leach: “Demand response experience in Europe: Policies, programmes and implementation”, Energy, 35(4), pp. 1575–1583, 2010, http://dx.doi.org/10.1016/j.energy.2009.05.021
Pierre Pinson, Henrik Madsen
Technical University of Denmark