IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 7, NO. 3, AUGUST 2011 381
Demand Side Management: Demand Response,
Intelligent Energy Systems, and Smart Loads
Peter Palensky, Senior Member, IEEE, and Dietmar Dietrich, Senior Member, IEEE
Abstract—Energy management means to optimize one of the
most complex and important technical creations that we know: the
energy system. While there is plenty of experience in optimizing
energy generation and distribution, it is the demand side that
receives increasing attention by research and industry. Demand
Side Management (DSM) is a portfolio of measures to improve
the energy system at the side of consumption. It ranges from
improving energy efficiency by using better materials, over smart
energy tariffs with incentives for certain consumption patterns, up
to sophisticated real-time control of distributed energy resources.
This paper gives an overview and a taxonomy for DSM, analyzes
the various types of DSM, and gives an outlook on the latest
demonstration projects in this domain.
Index Terms—Building automation, demand response, demand
side management (DSM), energy efficiency, energy management,
IEC 61850, load management, peak shaving, smart grids.
I. INTRODUCTION
T
HE CLASSICAL modus operandi of electric energy sys-
tems is unidirectional and top-down oriented. A limited
number of large power plants feed into the grid and try to keep
demand and supply balanced at all times. This balance is a very
crucial aspect in operating an electric energy system. Volatile
renewable energy sources [1] and electro-mobility are new
challenges to this balance and call for sophisticated control
methods [2].
Using the load as an additional degree of freedom is not en-
tirely new but affordable global communication infrastructure
and embedded systems make it now relatively easy to add a cer-
tain portion of “smart” to the loads. The development is driven
by the fact that—despite increased efficiency of electric de-
vices—consumption is steadily rising some percent every year.
While generation might not be much of a problem, it is the grid
capacity that makes many involved people worry.
Especially new and ambitious projects, like DeserTec (exten-
sive solar power stations in Northern Africa to supply Europe),
and large offshore wind parks in the Northern Sea, raise ques-
tions about the transport of energy. The grids might soon face
their limits, and intelligent Demand Side Management (DSM) is
one way to stretch these limits a bit further. DSM also promotes
Manuscript received February 27, 2011; revised May 05, 2011; accepted May
18, 2011. Date of publication June 27, 2011; date of current version August 10,
2011. Paper no. TII-11-080.
P. Palensky is with the Austrian Institute of Technology, Energy Department,
1210 Vienna, Austria (e-mail: palensky@ieee.org).
D. Dietrich is with the Vienna University of Technology, 1040 Vienna,
Austria (e-mail: dietrich@ict.tuwien.ac.at).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TII.2011.2158841
Fig. 1. Categories of DSM.
distributed generation: In order to avoid long-distance transport,
locally generated energy could be consumed by local loads, im-
mediately when it is available. DSM’s main advantage is that it
is less expensive to intelligently influence a load, than to build
a new power plant or install some electric storage device.
DSM includes everything that is done on the demand side of
an energy system, ranging from exchanging old incandescent
light bulbs to compact fluorescent lights (CFLs) up to installing
a sophisticated dynamic load management system. While DSM
was “utility driven” in the past, it might move a bit towards a
“customer driven” activity in the near future.
Reference [3] shows the utility point-of-view. The authors
perform a sequential Monte Carlo simulation to assess the im-
pact of stochastic grid component outages and how far DSM can
help in these cases. The correlation and sensitivity of the com-
ponent capacity variation to the expected shortage of available
transmission capacity is identified as well as the contribution of
DSM to transmission capacity. Such centralized structures are
sometimes complemented (if not replaced) by flat and freely or-
ganized market-driven mechanisms [4].
Depending on the timing and the impact of the applied mea-
sures on the customer process, DSM can be categorized into the
following (see Fig. 1).
a) Energy Efficiency (EE).
b) Time of Use (TOU).
c) Demand Response (DR).
d) Spinning Reserve (SR).
The “quicker” changes are processed and done, the more un-
wanted impact they potentially have onto the customers’ pro-
cesses. The “processes” can be manufacturing output, pump
power or even optimizing human comfort or health in a building.
1551-3203/$26.00 © 2011 IEEE
382 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 7, NO. 3, AUGUST 2011
Fig. 2. Impact of improved energy efficiency versus demand response.
The lower edge of the DSM spectrum is energy efficiency mea-
sures. They include all permanent changes on equipment (e.g.,
exchanging an inefficient ventilation system with a better one)
or improvements on the physical properties of the system (e.g.,
investing in the building shell by adding additional insulation).
Such measures result in immediate and permanent energy and
emissions savings and are therefore the most welcome method.
Sometimes seen as a separate category of DSM, Energy Conser-
vation (EC, [5]) shall be seen as part of Energy Efficiency in this
paper. EC focuses on users and behavioral changes to achieve
more efficient energy usage.
Time of use tariffs penalize certain periods of time (e.g.,
17:00–19:00) with a higher price, so customers (re)arrange
their processes to minimize costs. A change in the TOU
price-schedule means a change in a supply contract/tariff and,
therefore, does not happen on a frequent basis. [6] shows
that combining DSM and TOU tariffs significantly increases
security and lowers costs and emissions of energy systems with
a high share of wind power.
Dynamic DSM does not necessarily reduce energy consump-
tion, only consumption patterns are influenced. If a process is in-
terrupted for some reason, it might have to “catch up” once it has
green lights again. An example is a water pumping system that
can—because of its storage tanks—easily be shed for 30 min.
After the shed it has to fill up its tanks again, since they were
drained during the shed period. A so-called “rebound effect” (or
payback) takes place, energy is typically not saved and maybe
even a new peak is generated (see Fig. 2).
Sometimes this effect can be avoided, but it might result in a
reduced process quality. Such an ideal “peak shaving” applied
to a ventilation system would mean that if it normally ran at 50%
and were shed for half an hour, it is prohibited to compensate
that downtime with half an hour of 100%.
Reference [7] distinguish between the following.
a)
Incentive-Based DR.
• Direct load control (DLC): utility or grid operator gets
free access to customer processes.
• Interruptible/curtailable rates: customers get special
contract with limited sheds.
• Emergency demand response programs: voluntary
response to emergency signals.
• Capacity market programs: customers guarantee to
pitch in when the grid is in need.
• Demand bidding programs: customers can bid for
curtailing at attractive prices
b) Time-Based Rates DR.
• Time-of-use rates: a static price schedule is applied.
• Critical peak pricing: a less predetermined variant of
TOU.
• Real-time pricing (RTP): wholesale market prices are
forwarded to end customers.
Reference [8] distinguishes the following.
• Level I: Load shape objective.
• Level II: End use, technology alternatives, and market im-
plementation methods.
An alternative way to look at the various flavors of demand
response is to distinguish between the following.
• Market DR: real-time pricing, price signals and incen-
tives, and
• Physical DR: grid management and emergency signals.
Market DR relies on certain market places where prices are
formed and products are traded. Such market places are not ar-
bitrarily quick, which is why most transactions are done a day
ahead. An exception is real-time pricing (RTP), where the fig-
ures of an energy spot market (e.g., EEX—European Energy
Exchange in Leipzig) are forwarded to end users without delay.
A typical way to analyze and optimize market mechanisms is
System Dynamics (SDs). Reference [9] shows an SD approach
for designing energy prices for DSM. The price is split into
a capacity and a quantity part, a government-driven policy is
derived.
Reflecting grid congestion or an excess supply of wind power
onto the price can provoke stabilizing customer behavior. This
misleads some people to believe that monetary incentives like
RTP could solve all existing problems of the energy grid. How-
ever, limited customer elasticity and physical situations that are
not mapped onto prices lead to the fact that real load shedding
for grid relief cannot be done via prices alone.
This is where physical DR comes into play. It sends out
binding requests for demand management if the grid or parts of
its infrastructure (power lines, transformers, substations, etc.)
are in a reduced performance due to maintenance or failure.
A good mixture of both market and physical DR is usually
necessary to run a grid optimal.
Spinning Reserves (SR), implemented by loads, represent the
upper (i.e., quick) end of the DSM spectrum. Unfortunately, the
term itself is used in a loose way across the power community.
In this paper, spinning reserve is seen as primary (active power
output directly depends on frequency) and secondary control
(restoring frequency and grid state with additional active power)
[10]. This is typically the task of regulation power plants. Loads
can act as “virtual” (or negative) spinning reserve if they corre-
late their power consumption to the grid state in a “droop con-
trol” or some other smart manner. In its easiest way, devices use
less power if frequency drops (Fig. 3).
This can happen in an autonomous way (similar to primary
control) or in a coordinated way (similar to secondary control).
Depending on the type of DSM, different means of technology
and especially communication are necessary.
Comparing the various flavors of DSM, it is clear that EE is
most wanted. It saves energy and emissions, while most of the
PALENSKY AND DIETRICH: DEMAND SIDE MANAGEMENT: DEMAND RESPONSE, INTELLIGENT ENERGY SYSTEMS, AND SMART LOADS 383
Fig. 3. A cooperative consumer backs off when grid frequency decreases.
Fig. 4. A web-based energy information system, based on [11].
other types just shift it in time. So, the first goal must always
be to improve efficiency. After that, the dynamics can be op-
timized. Depending on the boundary conditions (electro-tech-
nical setting, market setting, capabilities of the system) of the
system, one or the other dynamic DSM method might be chosen.
Naturally, the break-even point of automation investments and
financial incentives determines how far one can go.
II. D
EMAND
SIDE MANAGEMENT
A. Energy Efficiency
Improving energy efficiency of buildings or industrial sites
starts with information and insight into the processes involved.
Practically, every customer site has hidden problems that waste
energy: compressed air leakages, misconfigured controls, dirty
filters, broken equipment, etc. Actually, such trivial problems
are often overlooked, unless a tool for analyzing energy effi-
ciency like [11] is used. The typical parts of such an Energy
Information System (EIS, Fig. 4) are as follows.
• Data acquisition infrastructure (sensor networks, data log-
gers, gateways, modems, etc.).
• An application server with database, calculation and anal-
ysis algorithms, alarming and reporting.
• User interfaces for visualization and configuration.
The classical calculations are as follows.
•
Baseline versus peak load comparison: a high baseline
might stem from standby power or old equipment (e.g., bad
insulation).
• Weekly comparison of time series: often lighting or venti-
lation accidentally runs through the night and the weekend.
Fig. 5. An energy controller switches (off) devices.
• Benchmarks: compare your performance to others, es-
pecially useful for multisite customers (e.g., supermarket
chains).
• Process correlations: does your energy consumption
strongly correlate with, for instance, the outside tempera-
ture or solar gains?
Beside such static efficiency figures, it is also the dynamics
that gets evident with such systems. An experienced facility
manager or a smart algorithm can interpret consumption pat-
terns and find ways to reduce peak loads. If the energy supply
contract penalized peaks, this would be a valuable result. If
changes to the logistics cannot help, automation equipment
might be necessary.
B. Energy Controllers
If the operations of equipment needs consumption-driven ad-
justment, an energy controller could be used. Such a device is
typically located at the energy meter and monitors the consump-
tion trend. If the trend points to unwanted levels, the controller
switches off equipment, based on certain priorities and other
rules (Fig. 5).
Configuring such an energy controller can be a very complex
task. Especially if consumers are added or removed, the stability
depends on a wise choice of rules. One simple example of how
to determine the priority level depending on the consumption
trend is shown in Fig. 6. The graphic shows one “measurement
period” (country-dependent time of usually 15 or 30 min that
represents the smallest time period for billing purposes). Each
period, the trend starts from zero and moves monotonously up-
wards. The more power is consumed (bottom curve), the steeper
the energy consumption curve.
Once the consumption trajectory crosses one of the upper
threshold lines, certain classes (or groups, priorities) of con-
384 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 7, NO. 3, AUGUST 2011
Fig. 6. Selection of priorities in a maximum demand monitor.
sumers are switched off (or duty-cycled, etc.). Fig. 6 assumes
three device classes, c1 the most important ones, c3 the least
important ones. The trajectory starts steep since all devices are
allowed on initially. It is steeper than the ideal (dashed) curve
so it necessarily crosses “c3 off” which turns off the least
important ones as a first measure to flatten the curve. In this
example, this is not sufficient so it crosses “c2 off” after some
time and also class 2 devices are switched off: only category
1 devices are allowed on now. The resulting (too) flat curve
crosses “c1 on” which has no effect since c1 devices were al-
lowed on anyway. Crossing “c2 on” allows class 2 devices on
again and so forth. “all off” would result in
and a hori-
zontal energy trajectory. With such a system, the consumption
trajectory will (if physically possible) reach the goal at the end
of the period.
Note that the power consumption during
and is different
although both intervals say “allowed devices classes are 1 and
2.” This demonstrates the fact that devices in one category may
be switched on but do not have to, since they have their own,
independent controls and schedules.
The goal can also change from period to period. With this,
a given load chart can be followed, always assuming that the
process physically allows that.
Fig. 7. OpenADR clients and the system operator connect to the DRAS.
C. Demand Response
A much quicker response is provided by the many flavors of
Demand Response (DR). Typically, a signal is broadcast, e.g.,
by the distribution or transmission system operator (DSO/TSO).
This signal might contain a price or a command for load shed-
ding/shifting. The deadline is not necessarily instantaneous: the
signal might refer to a situation next day at 12:00 noon since
often grid emergencies can be anticipated.
Classical Direct Load Control (DLC) assumes that loads are
fully under control, i.e., they do what they are told to do. All
intelligence is expected in the controller, who ideally uses load
models to make reasonable decisions. Reference [12] used a sto-
chastic state-space model for loads and simulate an urban power
system. The results show savings both in costs and transport
losses.
One modern system for automated demand response,
OpenADR [13], [14], is developed by the leading research
group on DR, the Demand Response Research Center (DRRC)
at the Lawrence Berkeley National Laboratory. OpenADR is
an open specification and an open-source reference implemen-
tation of a distributed, client-server oriented DR infrastructure
with a publisher-subscriber model. Its main components are the
following (Fig. 7).
• Demand Response Automation Server (DRAS).
• DRAS Clients at the customers’ sites.
• The Internet as communication infrastructure.
The client side is often just a communication library, used
by controls manufacturers to make their product OpenADR-ca-
pable. Clients can subscribe to DR “programs,” like critical peak
pricing or demand bidding, and the DRAS serves as simple
market platform and subscription manager. It keeps a database
of the participating clients and the program that they are sub-
scribed to.
If, for instance, a utility or system operator issues an emer-
gency message to the DRAS, the server forwards the message
to all clients that participate in the “emergency program.” Trans-
actions need to be recorded by the DRAS since financial incen-
tives are connected with reacting to such events.
The above system is almost open-loop control, since nei-
ther load models nor online feedback is used. Reference [15]
combines ripple control
1
with a wide area phasor measurement
system, based on global positioning system (GPS) timestamps
used by distributed voltage/current measurement equipment.
The result of adding this feedback loop is wide area control
for energy systems. Assuming 10% of the loads as controllable
1
Broadcasting powerline signals at nighttime for electric storage stoves.
PALENSKY AND DIETRICH: DEMAND SIDE MANAGEMENT: DEMAND RESPONSE, INTELLIGENT ENERGY SYSTEMS, AND SMART LOADS 385
via ripple control, the authors estimate approximately 30%
savings in transmission corridor losses and approximately 40%
in control power savings.
Reference [16] analyzes the business impacts of DSM. Four
business models are analyzed and a district of 300 households
with three different types of electric devices are simulated: loads
with storage (e.g., boiler), shiftable loads (e.g., dishwasher), and
real electric storages (e.g., batteries). The focus is on the district
level, i.e., making a district more self-sustainable with regards
to power consumption, and is entirely market based. It is shown
that flexible loads can be very attractive for DSOs when used
for substation-level peak shaving.
D. Distributed Spinning Reserve
Distributed spinning reserve tries to support the traditional
providers of ancillary services by imitating their behavior. On
the demand side, this means that load can be reduced or in-
creased when the grid frequency drops or rises.
Two implementations of this scheme are the Integral
Resource Optimization Network (IRON) [17] and the “grid-
friendly controller” [18]. Both measure the frequency and react
on it. The difference is that the IRON box has an additional
communication interface (a GSM/3G modem) that allows
cooperative algorithms.
A simple example for such an add-on feature due to its com-
munication capabilities is fairness. If, for example, a number of
devices can shed their load, and already one of them shedding
is sufficient, it might always be the quickest that wins the mon-
etary incentive. With communication, it can be arranged that all
of them have their turn. Such coordination also contributes to
stability. Imagine a community of autonomous, distributed con-
trollers without communication. All of them reacting on grid
problems in the same manner is the perfect recipe for instabili-
ties. They will have do it one after another to avoid a too strong
reaction.
All this is distributed, but still classical control: The frequency
needs to drop so that controllers react, and it needs to be restored
so that they stop fixing it. A slightly more sophisticated version
goes into model-predictive controls. A device that has a load
model of itself might predict how much/long it can shed load
until it has to stop shedding for process reasons. Load models
are the second step towards stability. They give the answer to the
question of how strong the reaction to an (anticipated) problem
needs to be and who can provide it.
E. Demand Shifting
Load models are also used when demand needs to be shifted
to other times. If the weather and other forecasts predict a grid
emergency at 17:30 the next day, intelligent consumers can plan
ahead and—if their process allows it—do their tasks earlier or
later. Examples are precooling, producing for the stock, etc. Pro-
cesses that can be shifted typically belong to one of the fol-
lowing categories.
• Inert thermal processes (heating, cooling).
• Inert diffusion processes (ventilation, irrigation, etc.).
• Mass transport (pumps with tanks, conveyor belts, etc.).
• Logistics (schedules, dependencies, lunch-breaks, etc.).
Shifting load to a later point in time (i.e., postponing) is easy.
The load is shed at the critical time and the process has to catch
up later. Unfortunately, the process quality is not guaranteed: If
there are not enough products on stock or if the tank is almost
empty, the process might run into troubles during the shed time.
It is therefore better to move the peak before the shed time and
be prepared. For this, load models are needed. They predict how
long things can be turned off, how much it takes to fill the “vir-
tual storage,” and what it costs [19].
The “virtual storage” of Demand Shifting can of course be
enhanced by special means. [20] for instance add phase change
material to buildings with electric heating to increase the low
thermal inertia of the structure.
F. Loads as Virtual Storage Power Plants
Virtual Power Plants (VPPs) are a community of typically
smaller generation units (often renewable energy sources) that
appear as one power plant to the grid management [21]. The
(typically distributed) equipment needs to be controlled from a
central dispatch and management node, and modern SCADA
standards like IEC 61850 [22] are used to integrate the indi-
vidual parts.
A special case arises if these parts are loads. Loads cannot
generate, they can only act as virtual storage via load shifting.
Aggregating many of such loads leads to sizes that can par-
ticipate on power markets and compete with traditional elec-
tric storage [23]. Typically, aggregators use proprietary tech-
nology to do this, but IEC 61850 is a good candidate to enable
interoperability.
The most crucial point of VPPs is even more crucial with
load-based virtual storage power plants: guaranteed availability.
If the grid operator requests a certain amount of regulation
power it must be delivered. Unfortunately, many loads behave
in a stochastic manner. A customer process might just at that
time not be interruptible or its virtual storage might be empty.
For this, again, reliable load models are necessary so that the
VPP operator can keep its promises available.
G. Communication Protocols for Load Management
IEC 61850 is a standards series for substation automation.
It contains an addressing scheme for “Intelligent Electronic De-
vices” (IEDs) or Distributed Energy Resources (DERs) and their
properties and functions, an XML-based substation configura-
tion language, communication protocols for best-effort and real-
time transport, and much more. Modern manufacturers imple-
ment this technology in their latest power engineering prod-
ucts like distribution automation nodes or grid measurement and
diagnostics devices. It is popular for implementing VPPs and
gradually makes its way down to the distribution level.
Coming from the other edge of the system, home and building
automation systems move towards the energy business. Smart
domestic controls can cause significant energy savings, [24] re-
port 40% energy savings via enhanced lighting controls. The
current question is how to unify the various devices from a con-
trols perspective?
Several technology groups have published standards for en-
ergy services in the home. One of them is the Zigbee Smart En-
ergy Profile [25], an extensive document that describes a large
![Fig. 4. A web-based energy information system, based on [11].](/figures/fig-4-a-web-based-energy-information-system-based-on-11-3v1phhl0.webp)