In This Issue
Winter Issue of The Bridge on Frontiers of Engineering
December 25, 2021 Volume 51 Issue 4
The NAE’s Frontiers of Engineering symposium series forged ahead despite the challenges of the pandemic, with virtual and hybrid events in 2021. This issue features selected papers from early-career engineers reporting on new developments in a variety of areas.

Enabling the Operation of Future Grids Using New Tools in Control Theory and AI

Tuesday, January 4, 2022

Author: Johanna L. Mathieu

New tools in control theory, optimization theory, AI, and machine learning are being developed to enhance grid reliability.

The shift toward a more sustainable energy future has led to a number of critical challenges in how to reliably and efficiently operate the electric power grid. Supply and demand must be balanced at all times, but renewable energy resources such as wind and solar produce power only when the wind blows and the sun shines, not necessarily when it’s needed, and there is very little grid storage. And while some proposed pathways to sustainable energy systems call for electrification of resources that directly consume fossil fuels (e.g., vehicles, heating systems, and some industrial processes) along with decarbonization of the electricity grid, electrification greatly increases the load on the grid, which in many parts of the country and world is already operating at its limit.

Distributed Energy Resources

To accommodate more wind, solar, and load, power systems researchers are developing approaches to exploit the flexibility of distributed energy resources (DERs). DERs can help the grid by responding to electricity prices or local measurements of voltage or frequency. However, they can provide even more value if aggregated to provide a large grid resource, sometimes referred to as a virtual power plant (Pudjianto et al. 2007).

DERs include small-scale storage mainly from batteries; electric loads with flexible consumption patterns (e.g., electric vehicles, air conditioners, and water heaters); and renewables such as residential solar photovoltaic (PV) systems with control functionalities enabling curtailment and/or reactive power control.

Choosing an Architecture to Coordinate DERs

A key research question is, What is the most effective architecture for coordinating DERs? Specifically, how should the power system operator, utilities, electricity consumers, and, possibly, third-party companies like DER aggregators interact?

A market mechanism, like transactive energy (Kok and Widergren 2016), would have each of them participate in a market to buy or sell electricity. In contrast, a control-based approach might have them all sign contracts with aggregators specifying a prenegotiated amount of flexibility that the aggregators would control in real time to participate in electricity markets by ­buying/selling energy and/or offering ancillary services.

The best choice of architecture depends on the ­product or services the resources are providing, the time­scale of participation, and consumers’ willingness to participate and engage in real-time decision making.

Ensuring Supply/Demand Balance through DER Control

Wind and solar necessitate more balancing services to ensure supply/demand balance (Makarov et al. 2009). Existing balancing services operate on timescales of seconds to minutes. For these fast timescales, I would argue that a control approach makes the most sense as it can achieve a reliable response and does not require frequent consumer engagement. In contrast, eliciting consumer responses through changes in electricity ­prices or direct participation in markets can result in inconsistent responses and may even synchronize demand, leading to oscillations that could be destabilizing to the power system (Nazir and Hiskens 2018).

A common objection to control approaches is that they leave consumers without sufficient autonomy over their own DERs. But consumer preferences can be accommodated during contract negotiation in which consumers and aggregators agree on the type/amount of flexibility offered, compensation, rules for control, and opt-out possibilities.

Challenges Associated with DER Control

There are a number of significant challenges in aggregating and coordinating DERs through control approaches. Consumers may be hesitant to give control of their DERs to an aggregator because of concerns about privacy and, in the case of flexible loads, worries about potential impacts on their life (e.g., uncomfortable home temperatures, uncharged electric vehicles). Further, coordinating large numbers of DERs can require transmitting large amounts of data (potentially very fast/often) and solving very large and complex control problems.

Additionally, the flexibility of most DERs is inherently uncertain. For example, solar PV curtailment and reactive power control potential are a function of uncertain solar power production and load flexibility is a function of uncertain load usage.

Finally, coordinating DERs to provide a particular service can exacerbate other grid issues. For example, providing balancing services can induce over- or undervoltages in the distribution network if network power flows are not explicitly considered by the controller (Ross et al. 2019).

Principles for Coordinating DERs

What principles should be followed when coordinating DERs to provide fast-timescale grid services?

  1. Don’t annoy the consumers. Minimize the need for private information and keep the control as nondisruptive as possible (Callaway and Hiskens 2011). For example, cycle air conditioners only within each home’s normal temperature range (usually around 1° plus/minus the temperature setpoint).Keep things inexpensive. Minimize measurement and communication requirements and use scalable and computationally tractable approaches that enable coordination of large numbers (hundreds to tens of thousands) of DERs.
  2. Plan for uncertainty. Ensure that the approach works (or, at least, does not fail catastrophically) when renewables and/or load do not match their forecasts and when the communication network is slow or goes down.
  3. Do no harm. Ensure that the control approach does not induce new grid problems or negatively impact the DERs themselves.

Promising New Approaches

New tools in control theory, optimization theory, artificial intelligence, and machine learning are being leveraged and further developed by grid researchers to address the challenges of DER coordination. There are innumerable recent papers in this space and rather than give a broad summary, I highlight some specific promising approaches that tackle these challenges and align with some of the principles above.

Time-Varying Optimal Control

Researchers at the National Renewable Energy ­Laboratory (NREL) have developed an approach to coordinate an aggregation of DERs connected to the same feeder (i.e., a portion of the distribution grid served by the same substation) to provide fast balancing services while managing distribution network constraints (Dall’Anese et al. 2018).

Power system measurements are gathered by an aggregator and used to solve a time-varying optimization problem; the resulting control actions are sent to DERs, which change their power consumption/­production. The feeder’s total power consumption/production then tracks a signal (e.g., a frequency regulation signal) that varies on timescales of seconds.

NREL has demonstrated this approach in practice with Holy Cross Energy in a net zero affordable housing community in Basalt, Colorado (O’Neil 2019). Importantly, this approach directly addresses the fourth principle—do no harm—by explicitly considering the distribution network impacts of DER control.

“Packets” of Energy

Researchers at the University of Vermont (UVM) and their spin-off Packetized Energy are developing and deploying a bottom-up approach to DER coordination (Almassalkhi et al. 2018). In contrast to the NREL approach, which is top-down in the sense that the aggregator sends control commands to the DERs, the UVM approach requires DERs like flexible loads to request “packets” of energy (i.e., to turn on for a fixed duration) and the aggregator approves or denies those packets. The aggregator’s choice of how many packets to approve enables it to shape demand to provide a grid service.

Packets are anonymous and therefore consumer privacy is preserved, addressing the first principle: don’t annoy the consumers. Packetized Energy has deployed its technology for residential water heaters and energy storage.

Nondisruptive Load Coordination

In my research group, we are developing approaches that are computationally tractable, are completely nondisruptive to electricity consumers, and minimize real-time communication needs.

For example, we have developed a method to ­coordinate thermostatically controlled loads like air conditioners and water heaters to track balancing signals (Ledva et al. 2018a; Mathieu et al. 2013). These types of loads switch on and off to maintain a temperature ­within a narrow band; our approach switches on and off the loads at slightly different times than ­normal to maintain temperatures within existing bands. It leverages Markov models to compactly represent the ­dynamics of a large aggregation of loads, a simple predictive controller, and Kalman filters to estimate states that we do not want to measure and transmit in real time, thus both reducing costs and preserving consumer privacy.

We also use online learning to estimate the real-time power consumption of the controlled loads, which is needed as feedback to our control algorithm, based on whatever measurements are already available from the network and loads (Ledva et al. 2018b; Ledva and Mathieu 2020). The online learning approach is data driven but uses dynamical system models to give some structure to the problem.

With funding from ARPA-E we are testing our load coordination ideas in practice, to identify possible technical issues when load coordination is implemented at scale and develop control approaches to address them, with an overarching goal of establishing credibility for load coordination at scale. Specifically, we are exploring distribution network impacts, unwanted dynamical behaviors like synchronization/oscillations, and problems that may arise due to imperfect communication networks.

We are doing physical experiments at Los Alamos National Laboratory and field tests on 100 households in Austin, Texas, with our partner Pecan Street Inc. The University of California, Berkeley is also a partner.


DER coordination is a critical tool to address some key challenges posed by the integration of high penetrations of intermittent and uncertain renewable energy resources like wind and solar. Innovations from control, optimization, artificial intelligence, and machine learning are being used by grid researchers to develop DER coordination approaches that address the four principles identified above to make DER coordination effective, reliable, and practical. Beyond leveraging existing ­theory and algorithmic tools, grid researchers are developing new theory and tools as we identify problems that existing approaches do not address.


Almassalkhi M, Espinosa LD, Hines PD, Frolik J, Paudyal S, Amini M. 2018. Asynchronous coordination of distributed energy resources with packetized energy management. In: Energy Markets and Responsive Grids, ed. Meyn S, Samad T, Hiskens IA, Stoustrup J. New York: Springer, 333–61.

Callaway DS, Hiskens IA. 2010. Achieving controllability of electric loads. Proceedings of the IEEE 99(1):184–99.

Dall’Anese E, Guggilam SS, Simonetto A, Chen YC, Dhople SV. 2018. Optimal regulation of virtual power plants. IEEE Transactions on Power Systems 33(2):1868–81.

Kok K, Widergren S. 2016. A society of devices: Integrating intelligent distributed resources with transactive energy. IEEE Power and Energy Magazine 14(3):34–45.

Ledva GS, Mathieu JL. 2020. Separating feeder demand into components using substation, feeder, and smart meter ­measurements. IEEE Transactions on Smart Grid 11(4):3280–90.

Ledva GS, Vrettos E, Mastellone S, Andersson G, Mathieu JL. 2018a. Managing communication delays and model error in demand response for frequency regulation. IEEE Transactions on Power Systems 33(2):1299–308.

Ledva GS, Balzano L, Mathieu JL. 2018b. Real-time energy disaggregation of a distribution feeder’s demand using online learning. IEEE Transactions on Power Systems 33(5):4730–40.

Makarov YV, Loutan C, Ma J, De Mello P. 2009. Operational impacts of wind generation on California power systems. IEEE Transactions on Power Systems 24(2):1039–50.

Mathieu JL, Koch S, Callaway DS. 2013. State estimation and control of electric loads to manage real-time energy imbalance. IEEE Transactions on Power Systems 28(1):430–40.

Nazir MS, Hiskens IA. 2018. A dynamical systems approach to modeling and analysis of transactive energy coordination. IEEE Transactions on Power Systems 34(5):4060–70.

O’Neil C. 2019. Small Colorado utility sets national renewable electricity example using NREL algorithms. NREL News.­colorado- utility-sets-national-renewable-electricity-example-using- nrel-algorithms.html

Pudjianto D, Ramsay C, Strbac G. 2007. Virtual power plant and system integration of distributed energy resources. IET Renewable Power Generation 1(1):10–16.

Ross SC, Vuylsteke G, Mathieu JL. 2019. Effects of load-based frequency regulation on distribution network operation. IEEE Transactions on Power Systems 34(2):1569–78.

About the Author:Johanna Mathieu is an associate professor in the Department of Electrical Engineering and Computer Science at the University of Michigan.