IEEE Electrification - September 2021 - 78

New York. They began the development of the TAG in 2013,
and the first transaction was conducted in April 2016.
Every prosumer on the platform can freely trade energy
directly without relying on a third party.
Designing and Operation Methods
Designing of MEMGs
The designing problem for MEMGs is to determine the
capacity configuration of various energy sources based on
the load demand. The main challenge involves the uncertainties
from renewable energies and loads. Two kinds of
methods are widely utilized to address these uncertainties:
stochastic optimization and robust optimization.
Stochastic Programming
If an accurate probability distribution of the uncertainties
can be obtained, the stochastic programming method can
be utilized to solve the designing of an MEMG. The most
commonly used method is scenario-based optimization.
For a known probability distribution, many scenarios can
be generated using sampling methods, such as Monte
Carlo and Latin hypercube sampling. Then, the capacity
configuration can be obtained by maximizing the expectation
of profit under these scenarios. Another popular
method in stochastic programming is chance-constrained
optimization, in which some constraints are satisfied with
a certain probability. For example, the outage probability
should be lower than an acceptable threshold. Then, we
can solve a chance-constrained problem to determine the
optimal capacity configuration.
Robust Optimization
In many cases, an accurate probability distribution is difficult
to acquire. In conventional robust optimization, we
only need to know the range of uncertain parameters, such
Centralized
as the intervals of renewable generations and loads. The
designer intends to maximize the profit in the worst-case
realization of uncertainty. After solving a max-min problem,
the capacity configuration results are obtained, which
will ensure the desired performance as long as the uncertain
parameters reside in the designated uncertainty set. If
an inexact probability distribution is known, distributionally
robust optimization can be used. This optimization technique
considers a family of distributions near the
empirical distribution and optimizes the statistic performance
under the worst-case distribution. The result
obtained from a robust optimization model is conservative.
Nevertheless, this technique is worthwhile if the MEMG is
expected to be able to survive in a critical environment.
Operation of MEMGs
The operation problem is to determine the energy supply
and consumption of each MEMG to maintain the energy
balance in the system. We will introduce the operation of
MEMGs using spatial dimension and time scale. In the
spatial dimension, the operation could be conducted in
both centralized and distributed manners. The time scale
varies from days, hours, and several minutes to real time.
Dispatch Center
Distributed
Information
Exchange
Centralized Versus Distributed
Similar to a traditional power system, the MEMG also has
a hierarchical control structure, i.e., primary, secondary,
and tertiary control, which can be implemented in either a
centralized or a distributed manner, as illustrated in Figure
3. In a centralized method, a dispatch center is needed
to collect information from all of the MEMGs, compute the
command centrally, and send it back to them. This works
well if the number of MEMGs is small. However, with an
increasing amount of MEMGs, a centralized operation
faces great challenges, such as problems of single points
of failure, low response speed, and the risk of privacy
leakage, all of which highlight the need for a distributed
strategy. Unlike the centralized method, the distributed
control strategy requires no control center; communication
between neighbors is enough. In addition, the real
values of generation and load are not needed to exchange,
which is good for privacy preservation. Well-known distributed
algorithms include the consensus, alternating
direction method of multipliers, dual decomposition, and
primal-dual gradient methods.
Control
Control
Control
Control
Information Exchange
Figure 3. A diagram of centralized and distributed operation.
78
IEEE Electrification Magazine / SEPTEMBER 2021
Offline Versus Online
Prediction is the basis of designing a suitable operation
strategy. If the prediction is accurate, then an offline dispatch
strategy can be formulated. However, with a high
penetration of renewable energies, the operational state
of an MEMG changes rapidly because of the fast variation
of power injections. This is also very difficult to predict
accurately. In this context, online decision-making methodologies
have been drawing increasing attention. Then,
the gap between optimization and control should be
IEEE Electrification - September 2021

Table of Contents for the Digital Edition of IEEE Electrification - September 2021
