Smart Grid and Renewable Energy, 2009, 18–22
Published Online September 2009 (http://www.SciRP.org/journal/sgre/).
From Smart Grids to an Energy Internet: Assumptions,
Architectures and Requirements
ABSTRACT
Secure and reliable delivery of energy is essential to modern society. Achieving this goal is becoming more challenging
with increasing demand and declining resources. The ongoing restructuring of the rather old delivery infrastructure is
an attempt to improve its performance so that energy can be utilized with higher efficiency. Smart grids are an ad-
vanced concept with a number of unique features compared to their precedents, including early detection and self heal-
ing capabilities. An implementation of smart grids is an energy internet where energy flows from suppliers to customers
like data packets do in the Internet. Apparent benefits from an energy internet are its openness, robustness and reliabil-
ity. This paper uses electricity as an example to present some key assumptions and requirements for building the energy
internet. An example is presented.
Keywords: Energy Internet, Smart Grids, Anticipatory System, Price Elasticity, Online Learning
1. Introduction
The importance of energy and its delivery infrastructure
for humanity can never be overstated. The availability of
resources determines that massive generation of energy,
such as electricity, has to be centralized. While custom-
ers are highly distributed, an extremely sophisticated
transmission and distribution network is needed for en-
ergy delivery. As an example, the electric power grid of
the United State of America consists of more than 3,100
electric utilities operating more about 10,000 power plan-
ts, and 131 million customers consuming more than
3,500 billion kwh every day. In the middle, there are
about 157,000 miles of high voltage electric transmission
lines. And on top of it, there are policy makers and regu-
latory agencies. The top priority of this massive and su-
per complicated infrastructure is to make sure that elec-
tricity is available whenever it is needed. Because its
service is socritical the grid was highly regulated for
nearly a century.
The dilemma is that our current knowledge about com-
plex systems like the electric power grid does not enable
us to regulate it efficiently and reliably. Often a com-
promise has to be made between efficiency and reliabil-
ity. And when it happens, reliability always has the pri-
ority. Even with this assumption, reliability is not always
guaranteed. It has been estimated that annual loss due to
service interruption is about $80 billion for the US alone.
[1] The Summer 2003 blackout in Northeastern US was
an example. The cause for each interruption might be
different. But the principal reason is that generation and
delivery capacity fails to provide sufficient safety mar-
gins, which may be further reduced due to poor man-
agement and regulation. Investing on more generators
and better delivery systems (transmission and distribu-
tion) is an answer with some severe limitations. There is
an unavoidable and unprecedented increase on energy
demand. Consumers will have higher expectations for the
service, both on quality and quantity. On the other hand,
resources are limited and the investment on exploiting
them is a lengthy and expensive process. In this picture,
undoubtedly the safety margin will be greatly reduced in
the future and more service interruptions will occur. An
alternative approach is the so called “generating through
saving”, which is inspired by the observation that a large
portion of the energy was lost due to the inefficiency in
operations.
Viewed from another angle, an energy infrastructure as
a complex system is in a healthy and stable state only
when all its components (10,000 power plants, 131 mil-
lion customers, 157,000 miles of transmission lines, and
more) are appropriately configured. This is an extremely
difficult optimization problem without any analytical so-
* L. H. Tsoukalas is with the School of Nuclear Engineering, Purdue
University, West Lafayette, IN 47907 USA (e-mail: [e-mail]
due.edu).R. Gao is with the School of Nuclear Engineering, Purdue
University, West Lafayette, IN 47907 USA (e-mail: [e-mail].e
du).

From Smart Grids to an Energy Internet: Assumptions, Architectures and Requirements 19
lution. The best tool available is the multi-agent (or intel-
ligent agents) approach, where the search for solutions,
rather than the solutions themselves, is explicitly formu-
lated. A framework (agent environment) is developed for
components to interact with each other and some proto-
cols are imposed on their interactions. With an effective
agent environment and the right set of protocols, compo-
nents can reconfigure themselves adaptively in order to
survive in the system. The system then acquires a very
desirable characteristic, that is, self healing. Whenever the
system deviates from its optimal operating point, its
components automatically reconfigure themselves to cor-
rect the problem. The Internet is one example where all
these things take place. In this sense, an Internet-type
network is a favorable option for the next generation en-
ergy infrastructure. The new system can provide an un-
precedented degree of flexibility to users, services pro-
viders, marketers, and regulators. Customers will be able
to choose the service package that fits their budget and
preferences thanks to competition. Service providers will
see more profits through organized production, a result of
real-time interactions with customers. Marketers or bro-
kers will have more information to plan more
user-oriented marketing strategy. The regulation agency
can operate to its maximal capacity by focusing effec-
tively on mostly on regulating issues. Best of all, it will be
a more reliable and more efficient energy infrastructure.
Unfortunately, the existing energy infrastructure is not
immediately ready for upgrading to an energy internet.
For example, most components, including millions of
electricity meters, in the current electric power grid are
passive with very limited communication and reconfigu-
ration capability. Furthermore, necessary regulations are
not in place for opening up the whole infrastructure.
Researchers have recognized the gap and tremendous
progress has been made on smart grids. [2] Smart grids
are an advanced concept with a number of unique fea-
tures compared to their precedents, (1) detecting and cor-
recting incipient problems at their very early stage; (2)
receiving and responding broader range of information;
(3) possessing rapid recovery capability; (4) adapting to
changes and reconfiguring accordingly; (4) building in
reliability and security from design, and (5) providing
operators advanced visualization aids. [3]
Progress made on smart grids has enabled new and
meaningful discussions on a full scale energy internet.
Building such a network requires substantial amount of
effort from diverse sectors such as technology, social
science, and legislation. Even though the Internet is a
full-fledged technology, some key differences between
energy (especially electricity) and electric data, which is
transmitted on the Internet, prevent a direct copy. They
are as follows:
1) Compared to electric data, electricity is mainly gen-
erated centrally and consumed locally. Long distance
transmission is critical and traffic control becomes im-
portant since routing options are usually limited. Bottle-
necks are more likely created.
2) Electricity cannot be stored at a large scale, which is
unique compared to the Internet where data are stored
and retransmitted. Storage, served as buffers, is an im-
portant stabling factor in a complex system. The lack of
storage in the electric power grid makes it vulnerable to
all kinds of instabilities.
3) The Internet uses a “Best-Effort” service model and
the quality of service (QoS) is a secondary consideration.
The energy network, however, assumes the opposite. The
top priority for the service network is to satisfy custom-
ers’ demand anytime. Therefore, for the Internet the
problem is how to allocate the bandwidth so that data
packets can be delivered efficiently. On the other hand,
in energy networks, customers’ peak demand, which can
occur at any time, will be closely monitored and fore-
casted so that generation/transmission/distribution can be
scheduled to meet this demand.
Recognition of the differences is a necessary step to-
wards a feasible energy internet. Of many possible solu-
tions to address these differences, the anticipatory control
methodology appears very promising. Anticipatory con-
trol is a set of tools that make control actions based on
system’s projected states.
The power of this paradigm comes from its farther fu-
ture vision as compare to conventional approaches which
use only current-state information to affect change. An-
ticipatory control consists of two parts, anticipation of
future states and intelligent decision-making based on the
anticipation. These two components, as discussed later,
are the key to fill the gap.
The rest of the paper will discuss an approach for
building the energy internet. The discussions include key
assumptions, critical requirements and candidate archi-
tecture.
2. A ssumptions
2.1 Intelligent Management and Sharing of
Information Achieve Virtual Energy Storage
As we have discussed before, the biggest obstacle for the
energy internet is the lack of significant energy storage
capacity inside the network. Without any feasible solu-
tion in the horizon, some researchers argue that it can be
virtually achieved via intelligent information manage-
ment and sharing. [4] The idea is to create a virtual en-
ergy buffer between customers and suppliers.
A virtual buffer is implemented through a demand side
management strategy which is build upon the practice of
dynamic data driven paradigms. With the emergence of
intelligent meters, it is possible to dynamically schedule
the use of electricity of every customer. This dynamic

From Smart Grids to an Energy Internet: Assumptions, Architectures and Requirements 20
scheduling will create a sheet of virtual buffer between
generation and consumption as we argue here, as shown
in Figure 1.
Under the new paradigm, the consumption of electric-
ity of every customer is intelligently managed. Custom-
ers don’t power up their electricity-hungry machines at
will. Rather, they make smart decisions after balancing
the costs and benefits. For example, some non-urgent
Figure 1. Virtual storage of energy
activities, such as laundry, can be scheduled for some-
time during the day or night when electricity is abundant
and cheap. The costs of the electricity are determined by
the supply-to-demand ratio and the capacity of the net-
work to transfer the resources. This managed use of re-
sources is analogous to the access control widely used in
the Internet. A buffer between generation and consump-
tion is therefore created, virtually. No physic laws are
broken. The electricity is still actually consumed when
generated. However, from the customer point of view,
with dynamic consumption scheduling the resources
(electricity) are created and then stored somewhere in the
power grid before they are used. The analogy is shown in
Figure 2. The virtual buffer may greatly increase the sta-
bility of the power grid.
Dynamic scheduling has to be carried out by software
agents or intelligent agents. The intelligent agents will
act on behalf of their clients; making reasonable decision
based the analysis of the situation. One of the most im-
portant analysis powers an agent has to possess is the
anticipation capability. The intelligent agent needs to
predict its client's future consumption pattern to make
scheduling possible. In other words, load forecasting
capability is the central piece for such a system.
2.2 Price Elasticity Can Effectively Manage
the Uncertainty
As presented earlier, prediction is the cornerstone of the
energy internet. However, accuracy is one of the major
concerns for using prediction data as the basis for gen-
eration. Uncertainty is always associated with predictions
and uncertainty may grow to an unacceptable level when
millions of predictions are summed up. To circumvent
this problem, a second assumption is presented, which is
that price elasticity can be used to effectively manage
uncertainty.
This assumption is analogous to the one used in feed-
back controls systems. Measurements are fed back as
references, which the controller can use to adjust its out-
puts adaptively. As a result the controller itself needs not
to be very accurate. Similar conclusion can be drawn here.
Provided a feedback loop between customers and suppli-
ers, prediction errors will be corrected adaptively. The
best feedback mechanism is provided by price elasticity,
particularly short-term elasticity. The mechanism of sh-
ort-term price elasticity will be discussed in more details
in the following sections. A good short-term price elas-
Figure 2. Internet and power grid with virtual buffers
ticity model provides an estimate on the customer’s pur-
chasing willingness with respect to the change of the
price. Through this tool, customers and suppliers can
perform dynamical negotiations to achieve a delicate
balance between generation and consumption, even with
less accurate load forecasting.
3. Architectures
There are many possible architectural candidates for the
energy internet as long as they satisfy the abovemen-
tioned assumptions. In this paper example architecture
will be delineated and discussed. The work presented
here was conducted mostly the Consortium for Intelligent
Management of Electric-power Grid (CIMEG).
In 1999, EPRI and DOD funded the Consortium for the
Intelligent Management of the Electric Power Grid (CI-
MEG) to develop intelligent approaches to defend power
systems against potential threats [5]. CIMEG was led by
Purdue University and included partners from The Uni-
versity of Tennessee, Fisk University, TVA and ComEd
(now Exelon). CIMEG advanced an anticipatory control
paradigm with which power systems can act proactively
based on early perceptions of potential threats. It uses a
bottom-up approach to circumvent the technical difficulty
of defining the global health of a power system at the top
level, as shown in Figure 3. The concept of Local Area
Grid (LAG) is extremely important in CIMEG. A LAG is
a demand-based autonomous entity consisting of an ap-
propriated mixture of different customers, charged with
the necessary authority to maintain its own health by
regulating or curtailing the power consumption of its

From Smart Grids to an Energy Internet: Assumptions, Architectures and Requirements 21
members. Once all LAGs have achieved security, the
whole grid, which is constructed by a number of LAGs in
a hierarchical manner, achieves security as well. To pur-
sue the health of a LAG, intelligent agents are used. Intel-
ligent agents monitor every load within the LAG, fore-
casting the power consumption of each individual load
and taking anticipatory actions to prevent potential cas-
cade of faults. A prototypical system called the Trans-
Figure 3. CIMEG’s implementation of an energy internet
mission Entities with Learning-capabilities and On-line
Self-healing (TELOS) has been developed and imple-
mented in the service area of Exelon-Chicago and Ar-
gonne National Laboratory. [6]
CIMEG’s mission was to create a platform for man-
aging the power grid intelligently. In CIMEG’s vision,
customers play more active role than in current power
system infrastructure. Lots of solicitation and negotia-
tions are involved, as shown in Figuer 4. The customer,
who is represented by an intelligent meter in Figure 4,
predicts a future need for electricity and places an order
in the market. The amount of the order is influenced by
the market price of the electricity, which is further de-
termined by the difference of the demand and supply and
also the capacity of the network. Economic models with
price elasticity are used in the process. [7] The active
interaction between customers and suppliers create a vir-
tual buffer between consumption and generation as dis-
cussed earlier.
4. Requirements
There are many possibilities to build an energy internet.
However, to make it successful, some requirements have
to be satisfied. It is out of the scope of this paper to enu-
merate the whole list. Instead, only some of the most
important ones are discussed here.
4.1 Smart Meter with Unique Address and
Communication Capability
This is probably the most fundamental requirement on
the side of hardware. Everything will start from a meter
installed on the customer side. The interactions between
customers and suppliers (including any middlemen, such
Figuer 4. Interactions among agents
as retailers) occur in such a high rate that manual opera-
tion is impossible. Basic hardware support is needed for
automatic and real-time communication between cus-
tomers and suppliers. That necessitates a smart meter
with a unique and addressable identifier and two-way
communication capability.
4.2 Forecasting Capability
Prediction and anticipation are the essential stabling
forces in a complex system. The more information about
the future is known, the better planning can be made. For
an energy internet, it is crucial that customers’ energy
usage patterns can be predicted with a degree of certainty.
There are plenty of tools available for this purpose thanks
to extensive research efforts in the past. Using parametric
(statistical) or non-parametric (neural networks) or hy-
brid (fuzzy logic) methods, these tool can predict very
accurately a customer’s short-term demand. [8]
4.3 Multi-Resolution Agents
The real time operation in an energy internet requires the
application of intelligent agents. Intelligent agents act on
behalf of their clients, who can be actual customers,
power grid operators, electricity brokers, or non-human
entities such as transformers, generators, transmission
lines. Intelligent agents are equipped with sufficient
knowledge so that they can act rationally. Conventional
wisdom suggests that each intelligent agent should take
actions to pursue maximal benefit for itself. This as-
sumption made to simplify the operation of intelligent
agents. However, it may result in some unwanted side
effects that have been identified by many researchers

From Smart Grids to an Energy Internet: Assumptions, Architectures and Requirements 22
especially in game theory. These side effects are harmful
to the health of the whole system and appropriate actions
should be taken to avoid that.
In classic game theory, an agent is given access to all
information if available, such as the possible actions and
outcomes of other agents. We argue that this is an unre-
alistic condition. In classic game theory, an agent is able
to choose the best strategy (i.e., act rationally) if it can
examine all possible scenarios. Experimental data con-
tradicts such conclusions (see Prisoner’s Dilemma). Hu-
man beings sometimes act irrationally according to game
theory. A human being does anticipatory reasoning,
which means that when making a decision the conse-
quence has been taken into consideration. The major dif-
ference with classic theory is that human being does not
have an accurate prediction for things in very far future,
which in part explains why a human being sometimes
acts irrationally according to classic theory. A human
being sees the future in two different time scales, short
term and long term. In the short term scope, we can make
very accurate estimations. In long term scope, we have to
make estimations with increasing degree of uncertainty.
Therefore, a more realistic agent must possess a multi-re-
solution vision. We propose it as a second principle (as-
sumption) for agents. We shall show that this is not only
a reality but also a stabling factor for complex systems.
In classical game theory, agents compete for their max-
imal benefits (payoffs). This assumption excludes them
from collaboration in many situations (such as Prisoner’s
Dilemma). For agents with multi-resolution vision, col-
laborations are possible because the future is not clearly
mathematically defined. For example, in the Centipede
game, if both players know exactly what the other pla-
yer’s move (future), the first player is likely to choose to
defect on the first play because this is Nash equilibrium
and he has no incentive to choose the other option (coop-
eration). However, in reality, the future that players can
see is not a clear picture. But the most important informa-
tion a player can read from this fuzzy picture is that the
payoffs would be much better in the future. A player is
likely to choose to cooperate if he knows he can gain
more by passing the piles. Then he would be very likely
to choose cooperate in order to maximize the personal
gain.
4.4 Short-Term Price Elasticity Model
Price elasticity is used to characterize the sensitivity of
customers to the change of the price. In the case of elec-
tricity, the price elasticity measures how the price change
impacts the customers’ willingness to consume power. A
good short-term price elasticity model provides the basis
for interactions between customers and suppliers.
5. Conclusions
Building an internet type of energy network for the future
may be the answer to some of the pressing energy chal-
lenges. Advancements in information technology and
ongoing research on power infrastructure and complex
system has made this goal reachable. The paper summa-
rized some of the fundamental assumptions and require-
ments and presented an example architecture as well. The
discussion was focused on technical and marketing issues.
It is noteworthy that the subject requires inclusion of
work from several other areas including, economics,
regulation, resource management, and market structures
for and capital allocation and risk management.
REFERENCES
[1] Hamachi LaCommare, K., and J. Eto., Understanding the
Cost of Power Interruptions to U.S. Electricity Consum-
ers, LBNL–55718, http://certs.lbl.gov/pdf/55718.pdf,
September 2004.
[2] European Commission, European SmartGrids Technol-
ogy Plaform: Vision for Europe’s Electricity Networks of
the Future, EUR 22040, 2006.
[3] EPIC, San Diego Smart Grid Study Final Report, October
2006.
[4] Gao, R. and Tsoukalas, L. H., Implementing Virtual
Buffers for Electric Power Grids, ICCS2007, Beijing,
May 2007.
[5] CIMEG, Intelligent Management of the Power Grid: An
Anticipatory, Multi-Agent, High Performance Computing
Approach, EPRI, Palo Alto, CA, and Applied Intelligent
Systems Lab, School of Nuclear Engineering, Purdue Un-
iversity, West Lafayette, IN, 2004.
[6] Gao, R., Tsoukalas, L. H., Anticipatory Paradigm for
Modern Power System Protection, ISAP 2003, Lemnos,
Greece, August 31–September 3, 2003.
[7] Gao, R. and Tsoukalas, L. H., Short-Term Elasticities via
Intelligent Tools for Modern Power Systems, MedPow-
er’02, Athens, Greece, November 4–6, 2002.
[8] Tsoukalas, L., and R. Uhrig, Fuzzy and Neural Appro-
aches in Engineering. New York: Wiley, 1997.