IEEE Power & Energy Magazine - November/December 2020 - 44

1.05

Voltage Magnitude (p.u.)

1
0.95
0.9
0.85
Voltage Without Control
Voltage With Regulators
Voltage With OPF

0.8
0.75

1,000

2,000
3,000
Node Index on the Primary Side

4,000

figure 7. The voltages are controlled within 0.95-1.05 p.u. by the OPF algorithm (green dots).

figure 8. A fleet of EVs under distributed control in the
NODES experiment at NREL's Energy Systems Integration
Facility. (Source: NREL; used with permission.)

machine learning, scalable simulations, and data analytics-
are being employed to develop and evaluate these new control
and optimization algorithms at large scales to operate millions to hundreds of millions of controllable devices on the
grid in real time.
Innovative, secure, scalable, hierarchical, real-time control strategies that are autonomous and make the best use
of big and real-time streaming data will be explored to
ensure that these complex systems function properly under
a wide range of possible conditions. Evaluating deployments
through coordinated simulations of 10 8 devices, including
high-fidelity models of the system, each component (e.g., residential/commercial buildings, autonomous EVs, solar, wind),
and autonomous controllers in both normal and abnormal
44

ieee power & energy magazine

operations will be carried out to characterize and validate
these approaches.
Currently, it takes approximately 1.5 h to run a simulation
of feeders with 12,000 DER devices and optimization-based
distributed controllers, including high-fidelity solutions of ac
power flows for evaluating 24 h at 1-min resolution. This
equates to about 23 days of simulation to run an entire year.
Scaling to tens of millions of devices will require much
more for an annual simulation. In many cases, simulations
of critical days and weeks are sufficient; however, this highlights the need for advanced computational resources to
fully evaluate these control and optimization approaches
for the AEGs of the future. Luckily, in the future, once the
algorithms are developed and verified, simulations of their
outcomes will not be needed to run in real time on the grid.

Evaluations in the NREL Energy
Systems Integration Facility
To evaluate if the software algorithms would work when
integrating many real controllable devices, we set up a large
experiment at NREL's Energy Systems Integration Facility. NREL's work on the ARPA-E NODES program helped
develop the first implementation of the algorithms in hardware and successfully demonstrate the real-time optimization of a single AEG cell. The experiment included simulation of a real distribution feeder from California with 366
single-phase connection points, more than 100 controllable
assets at power (inverters, EVs, and batteries; see Figures 8
and 9), and hundreds of simulated devices. The distributed
algorithms were implemented in cost-effective microcontrollers that self-optimize and communicate to the central
coordinator to attain systemwide goals (voltage regulation
and frequency response).
november/december 2020

IEEE Power & Energy Magazine - November/December 2020

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - November/December 2020