IEEE Power & Energy Magazine - November/December 2020 - 39

Operate in Real Time

first step is to develop convex relaxations and linear approximations of pertinent nonconvex problems.
✔✔ Model inaccuracy. Approximate linear models or
convex relaxation methods might be leveraged to derive convex problems that facilitate the design of computationally affordable solutions. However, approximate/relaxed convex problems might involve only
an approximate representation of a system's physics
and constraints; therefore, the optimal solutions of the
convex problem might not be feasible for the original
problem. To begin to address this issue, distributed
optimization algorithms have been developed to use
measurement information directly, which is known
as online optimization with feedback. Measurementbased (or feedback-based) algorithms address the feasibility issue, and they can be distributed or centralized. The design of a distributed version is certainly
more challenging than the centralized one, but the distributed version can be implemented on a more flexible communications architecture, which can enhance
cyber robustness.

One key challenge with AEGs is the development and implementation of real-time optimization and control methods.
We use the term real time to indicate that power set points of
the DERs are updated within each cell on a second or subsecond timescale. Electric grids must maintain energy balance
at every time instance. This is required to maximize the
operational and economic objectives while coping with the
variability of ambient conditions and noncontrollable energy
assets and achieving intercell coordination to ensure reliable
systemwide operation. Solving optimization problems to convergence every second or every few seconds, however, has
been impractical because of the following challenges:
✔ ✔ Complexity and convergence analysis. For largescale grids, the computat iona l complex it y of a
centrally defined system could prevent the solution of
optimization problems at the required timescales. When
an optimization problem is solved in a distributed and/or
hierarchical fashion (e.g., with device-to-device or cellto-cell communications as well as intracellular message
passing), multiple communications rounds are necessary
to converge to (possibly optimal) solutions. Note that the
optimization tasks related to
AEGs are markedly different
100,000,000 s
from traditional settings in
which energy systems are optimized at the wholesale level
using economic- and market-based objectives. In the
traditional operation of bulk
1,000,000 s
systems, a few large-scale
generators are dispatched,
and the noncontrollable net
load varies slowly. Such operation is incompatible with
AEGs that include a massive
integration of DERs or whose
100,000 s
optimization models require
accurate representations of
ac power flows within the
DERs' controllability region.
In traditional bulk systems,
optimization problems are
nonconvex, nondeterministic,
1,000 s
and polynomial-time hard
(NP-hard); therefore, they
may be infeasible to solve
at the envisioned timescale
1-100 s Renewable Conventional
EV
Industrial Commercial Residential
with hundreds of millions
of control points. To address
these challenges and facilitate figure 2. The AEGs form a distributed hierarchical control system that integrates
the development of provably individual technologies in a cellular structure to the bulk power system. The scale
stable and optimal distributed on the side indicates the number of controllable technologies seen along the bottom
solution methods for AEGs, a level. The lowest level depicts the locations of various generation, storage, and loads.
november/december 2020

ieee power & energy magazine

IEEE Power & Energy Magazine - November/December 2020

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