By Sean Meyn
The power grid in the U.S. and many regions of the world is implementing new technologies in an effort to realize the “smart grid” vision. Smart meters and a smarter grid will likely lead to more efficient use of our infrastructure. In addition, increased renewable energy integration has the potential to provide power at low cost. This optimism may be justified, but only if experts in control theory play a leading role.
Within the next decade it is expected that renewable energy will be ubiquitous, exposing the grid to levels of volatility it has never experienced before. As consumers install more solar panels and energy storage devices, dynamic models of the grid will require revision.
Presented with so many changes, it is an exciting and daunting time to be in charge of the grid. But the challenges faced by tomorrow’s grid operator are manageable, as long as we keep in mind a basic lesson of engineering: our models of the grid and of human behavior are only approximations of reality.
Volatility from Renewables
Figure 1 shows power from wind generation in the Bonneville Power Administration (BPA) region in the Pacific Northwest during the first days of 2015. The vertical axis is measured in gigawatts. The first day of the year was completely calm, with virtually no energy from wind, yet the maximum power on January 5 is similar to the output of four typical nuclear reactors. In addition to this fluctuation in wind energy, the sun also brings volatility: California’s rush to increase solar energy penetration may mean that the net load will be near zero during lunchtime on sunny days. This is great news, in the sense that these energy sources are clean, domestic, and renewable. The question, however, is, how do we deal with this massive disturbance to the grid?
Figure 1. Wind generation during a typical week at BPA.
One of several proposed resources is batteries. Southern California Edison (SCE), which supplies electricity to much of Southern California, recently announced the “largest battery in the world.” The SCE battery is a 6,300 square-foot facility capable of storing 32 MWh (megawatt-hours) of energy. This is not a lot. It would take one hundred times this amount of storage to address the ramps seen on January 4 and 5 at BPA. In addition to their enormous investment cost and size, batteries do not “store for free.” They have a limited lifetime, and waste energy as they charge and discharge.
Consumers in the Loop
Fred Schweppe’s vision of a grid in which primary control would be augmented by the action of loads ramping power consumption up and down in response to local measurements inspired dreams of a better grid among academics . He and many others have also argued for real-time prices to consumers as a mechanism for real-time control, putting consumers in the loop and introducing new dynamics. Both market analysis  and control considerations cast doubt on this approach to regulating the grid.
This begs a multitude of questions. Do we expect consumers in Seattle to track the entirety or a portion of BPA wind generation? Where most of the loads are on/off devices, how can consumers be expected to track continuous regulation signals? Will uncertain consumer response, and delayed response, lead to an unstable feedback loop? What effect will this have on water heaters and refrigerators that are ramping up and down to service the grid? To answer these questions, we need to look more closely at the control problem faced by today’s grid operator.
Grid Control Loop
Figure 2 shows the grid from the viewpoint of a typical balancing authority (BA), where grid operators are in charge of maintaining stability and reliability in their region of the grid. The measurement \(Y(t)\) is a sum of error signals, including deviation in grid frequency (60 Hz in the U.S.). In most cases, the compensator \(G_c\) is proportional-integral (PI) control, designed by the grid operator to maintain supply-demand balance.
Figure 2. Power grid control loop.
The actuation block \(H\) models the cooperation of many ancillary services acting in parallel, provided by the electric grid in order to facilitate and ensure the continuous flow of electricity so that supply will continually meet demand. Each service receives a command from a BA and adjusts power up or down in response. Resources providing ancillary service include responsive generators and batteries.
The grid transfer function \(G_p\) takes as input power mismatch and output \(Y(t)\), as shown in Figure 2. We have many observations of the grid following a generation outage that can be used to construct a grid model. It is fortunate that these dynamics are quick, relative to disturbances from nature; the natural frequency in a second order model for \(G_p\) corresponds to timescales of less than 20 seconds . The disturbances entering the grid from renewable energy evolve on much slower timescales, where the grid transfer function \(G_p\) appears as a static gain.
When viewed as a centralized control problem, renewable energy integration does not appear to be a great challenge because grid dynamics are relatively tame on the timescales of disturbances introduced by solar or wind generation. The challenge lies in finding resources that can be harnessed to create the actuator block \(H\) in Figure 2. If this is to be achieved using flexible demand, we must investigate what is meant by flexibility.
While consumers expect some guarantees on quality of service (QoS), the grid operator at a BA desires reliable ancillary service obtained from inherent flexibility in the consumer’s power consumption. These seemingly-conflicting goals can be achieved simultaneously by local control; an appliance can monitor its QoS and other state variables, receive information from the BA, and adjust its power consumption accordingly. Methods to create “virtual energy storage” (VES) from flexible loads offer reliable resources to grid operators without impacting the needs of consumers.
Experiments conducted at the University of Florida have shown that loads in commercial HVAC systems can provide regulation at timescales of one to 20 minutes through manipulation of the fans in the ventilation system; a local feedback loop ensures that the power deviation provided to the grid tracks the desired regulation signal. The cost is negligible, with no impact on building climate . Installation is simple, because these commercial buildings are already equipped with building automation systems. The aggregate capacity of ancillary service obtained is on the order of gigawatts in this frequency range.
Many loads that can provide VES in lower frequency ranges consume power at discrete values. Figure 3 shows power consumption for a typical residential refrigerator. The lighter plot shows the power consumption deviation that would be observed using local control for VES. These minor variations in cycling are not expected to impact QoS to the refrigerator owner. But how can this be useful for the grid?
Figure 3. Nominal power consumption of a residential refrigerator is approximately periodic.
Through several layers of local control, a collection of many refrigerators can provide VES in a frequency range near the natural frequency of the loads (in this case, corresponding to a period of one hour). A randomized control architecture creates the degrees of freedom required to track the regulation signal received from the balancing authority . The aggregate of loads will appear as a perfect battery, subject to constraints on the frequency range of ancillary service it provides. Of course, residential refrigerators are only a tiny fraction of the flexible loads that can be harnessed in this way.
In conclusion, there is incredible value in placing consumers in the loop. This is only possible through careful design of local control loops to ensure that both the consumer and the grid receive what they need. Naturally, even if there is no cost or loss of QoS, the consumer needs incentives to participate. Contracts between BAs and ancillary service providers are highly successful today. Fixed payments for engagement and regular payments proportional to “services rendered” may incentivize more flexibility from consumers, leading to a more robust grid that can manage greater levels of renewable energy.
Power demand minus renewable generation
Acknowledgments: The research surveyed here was supported in part by NSF grants CPS-0931870 and CPS-1259040. Many thanks to my colleagues and to Bob Moye, who provided comments on the first draft of this article.
 Lin, Y., Barooah, P., Meyn, S., & Middelkoop, T. (2015). Experimental evaluation of frequency regulation from commercial building HVAC systems. IEEE Trans. on Smart Grid, 6(2), 776–783.
 Mathias, J., Kaddah, R., Bušić, A., & Meyn, S. (2015). Smart fridge / dumb grid? demand dispatch for the power grid of 2020. Proc. 49th Annual Hawaii International Conference on System Sciences (HICSS). To appear online at arXiv:1509.01531.
 Meyn, S., Barooah, P., Bušić, A., Chen, Y., & Ehren, J. (2015). Ancillary service to the grid using intelligent deferrable loads. IEEE Trans. Automat. Control, 60(11), 2847–2862.
 Robinson, S. (2005). Math model explains volatile prices in power markets. SIAM News, 38(8).
 Schweppe, F., Tabors, R., Kirtley, J., Outhred, H., Pickel, F. & Cox, A. (1980). Homeostatic utility control. IEEE Trans. on Power Apparatus and Systems, PAS-99(3), 1151–1163.