INL/CON-13-28641

PREPRINT

Using System Dynamics

to Define, Study, and
Implement Smart Control
Strategies on the Electric
Power Grid

The 31st International Conference of

the System Dynamics Society

Lyle G. Roybal
Robert F. Jeffers

July 2013

This is a preprint of a paper intended for publication in a journal or

proceedings. Since changes may be made before publication, this
preprint should not be cited or reproduced without permission of the
author. This document was prepared as an account of work
sponsored by an agency of the United States Government. Neither
the United States Government nor any agency thereof, or any of
their employees, makes any warranty, expressed or implied, or
assumes any legal liability or responsibility for any third party’s use,
or the results of such use, of any information, apparatus, product or
process disclosed in this report, or represents that its use by such
third party would not infringe privately owned rights. The views
expressed in this paper are not necessarily those of the United
States Government or the sponsoring agency.
 Using System Dynamics to Define, Study, and Implement Smart
Control Strategies on the Electric Power Grid

Lyle G. Roybal and Robert F. Jeffers

Abstract
The United States electric power grid is the most complex and expansive control system in the
world. Control of individual generators is based on unit inertia and governor characteristics,
larger regional control coordinates unit response based on unit economics and error conditions,
and higher level large-area regional control is administered by a network of humans guided by
economic and resiliency related factors. Under normal operating conditions, the grid is a
relatively slow moving entity that exhibits high inertia to outside stimuli, and behaves along
repeatable diurnal and seasonal patterns. However, that paradigm is changing due to increasing
implementation of renewable generation sources. Renewable generators by nature cannot be
tightly controlled or scheduled and appear like a negative load to the system with all of the
variability associated with load on a larger scale. In response, grid-reactive loads (i.e. smart
devices) can alter their consumption based on price or demand rules, thereby balancing this
variability. This paper demonstrates how a system dynamics modeling approach capable of
operating over multiple time scales can provide valuable insight into developing new “smart-
grid” control strategies and new ancillary services for smart devices to accommodate renewable
generation and regulate the frequency of the grid.

Key Words
Smart Grid, power systems, load management
Introduction
Electric utilities throughout the United States and the world are progressing towards higher
levels of intermittent and distributed renewable energy. With higher levels of intermittent
generation, power system planners and operators are struggling to provide the ancillary grid
services needed to integrate this renewable energy. This trend will increase the value of ancillary
services, especially those that respond on timescales similar to those of intermittent generation,
such as the frequency regulation service. From traditional energy storage systems alone,
revenues from ancillary services applications are forecasted to rise from $20 billion in 2012 to
between $30 and $55 billion by 2022 [1]. Large-scale grid energy storage has been heralded as
the breakthrough needed for economic integration of renewable energy sources. However,
utilities are currently finding large-scale grid storage technologies that can respond quickly
enough to integrate intermittent generation to be prohibitively expensive [2]. While battery-based
grid storage may reach a technology breakthrough in the next decade, the potential for
intelligently-controlled load (herein: load management) to provide these services should also be
considered as a readily-accessible option.

The current paradigm for considering load management as a resource does not adequately take
into account emerging technologies that would allow load to provide highly valuable frequency
1
regulation services. In fact, the behavior of traditional load at sub-hourly timescales is
underrepresented by the current state-of-the-art power system modeling tools. These modeling
shortcomings have lead to a misrepresentation of the potential for load management
technologies, which ultimately results in a very shallow innovation learning curve.

In order to shift toward a more aggressive learning curve that will change the way the industry
approaches load and lead to a robust capability to integrate renewable generation in a short
amount of time, we have developed a modeling tool based on system dynamics principals that
utilizes a multi-timescale modeling approach for power systems. This model is able to simulate
technology innovations and integrate economic impacts with power system operation constraints
over multiple relevant timescales. This has significant impact to the way the power industry
approaches ancillary services for frequency regulation.

Using this modeling tool, utilities will be able to model system-wide impact of load management
technologies on grid stability. Regulators will be able to collaborate with utilities to set new
pricing structures that capture the true value of new ancillary services. Policy-makers will be
able to utilize the simplified model interface to help them make legislative decisions regarding
the support of renewable energy and the impacts that load management devices will have on the
economy. By providing a deeper understanding to these high level issues, the electric industry
can eliminate roadblocks in the slow cycle of load management development, which will
dramatically decrease the cost of ancillary services critical to intermittent generation integration
and the growth of renewable energy.

National Needs and Problems Encountered

The United States has aggressively pursued renewable energy generation technologies to meet
the Presidential Directive [3] of increasing our energy security and to reduce the generation of
greenhouse gases. Other than the mature technologies of hydro-electric generation and pumped
hydro-electric generation, the leading candidates for large quantities of renewable generated
electricity are wind and photo-voltaic (PV) or solar farms. This mentality has filtered down to the
state level of implementation and the majority of the states in the United States have Renewable
Portfolio Standards (RPS) and goals for the implementation of renewable generators [4].

A renewable portfolio standard is a policy that requires electricity retailers to provide a minimum
percentage or quantity of their electricity supplies from renewable energy sources. An RPS
establishes a base level of demand but allows the market to determine which renewable energy
resources will meet that demand. Historically, state legislatures and regulatory agencies have
been the driving force behind RPS policy formulation, although some RPS polices have been
adopted through citizen ballot initiatives. Initially proposed as a mechanism to support renewable
energy development in competitively restructured electricity markets, the RPS model today
serves additional policy aims such as fuel diversity and in-state economic development. By the
end of 2007, 25 states and the District of Columbia had enacted RPS policies, ranging from 2%
of the electricity supply in Iowa to 40% in Maine (Figure 1). Three other states, Illinois, Virginia
and Vermont, have established nonbinding renewable energy goals. The time horizon for
achieving the RPS varies among states.

2
Figure 1. RPS goals and participation by state [4].

The nature of electric power is that whatever is generated must either be consumed immediately
or stored for later use. One mature method of storage is pumped hydro-electric systems where
water is pumped to a higher potential energy during times of excess generation for use at a later
time. While this technology is viable, suitable sites are limited and capital costs are high.
Battery storage is a topic of intense research, but remains an expensive and real-estate intensive
option. So, currently most power systems in the world operate under the premise that generation
and consumption must be matched nearly instantaneously. This is done by close monitoring of
power system frequency and voltage, accurate load forecasting, and control of generation output
to match current load conditions.

In reality however, there are always minor imbalances in the generation and consumption. This is
because consumption, or load, has a random element to it that is unpredictable within a given
range or percentage of total load. Generation/load mismatches are absorbed by the rotating
machinery of the generators in the form of kinetic energy. Too much generation causes the
generators to speed up, increasing the frequency of the generated electricity. Too little generation
(or too much load) causes the generators to slow down, decreasing the frequency of the system.
Ideally, frequency of the system is maintained within one-tenth of percentage point of 60 Hz in
the United States to protect sensitive electronics and provide adequate power quality.

Unfortunately, wind and PV renewable generators can and often have a large random component
to their output. This is the nature of wind and PV as the wind can change speed and direction
randomly, and clouds can occlude PV farms in a random fashion. Therefore, renewable
generation appears to the electric utility as a negative load owing to its randomness. They can
also have a much larger random component compared to normal loads as illustrated in figure 2

3
[5]. The left plot in Figure 2 demonstrates how quickly significant wind generation can be lost.
At approximately 6:00 AM, a total of 400 MW of wind generation is lost in a matter of minutes.
Because this loss of generation is likely not forecasted, the utility must hold extra capacity online
as spinning reserve (online generators operating below their maximum output) to absorb these
random fluctuations. This capacity is sometimes referred to as incremental reserve. The right plot
in Figure 2 demonstrates that wind can also increase output quickly, requiring the utility to scrub
generation from responsive plants. Utilities must have machines online that can quickly decrease
their output, termed decremental reserve. Ultimately, the inefficiency of this operation costs
both valuable natural resources and money.

Figure 2. Wind profiles showing dramatic decrease and increase of overall wind generation
during a day [5].

Solar power ramps tend to be even faster than the wind ramps shown here. Frequency regulation
is an ancillary service provided by incremental and decremental reserves that are able to respond
on timescales of a few seconds to minutes. It is so-called because the electrical frequency of the
system is proportional to the amount of kinetic energy stored in the rotating masses of
generators, and is therefore the integral of net power imbalance over time. On timescales of
minutes to hours, the service of slower-responding reserves is marketed as load following. Fast-
ramping frequency regulation is the most valuable ancillary service because it is the most
expensive to provide. This is because fast-responding plants such as gas turbines nominally have
higher operating costs than slow-responding plants such as the common coal and nuclear
technologies. Grid storage technologies such as advanced batteries are theoretically excellent
frequency regulation providers because they offer both quick response and precise control.
However, to provide frequency regulation at the scale required for large utilities, advanced
battery storage remains very expensive [6].
Frequency regulation services required to maintain sub-second energy balance in the electric grid
have traditionally been supplied by generators and are often procured as an ancillary service in
organized markets. However as the imbalances increase in both frequency and amplitude due to
renewable generation, it becomes very expensive to do this with traditional generation units.
However, this can be accomplished more efficiently and economically using intelligent load
management techniques [7]. The model developed herein aims to provide a method and
framework by which frequency regulation schemes can be postulated and analyzed using
distributed load management techniques incorporating the behavioral aggregate of smart homes

4
and devices while including traditional load following and other ancillary frequency regulation
services.
The Need for New Power Modeling Techniques – Top Level Design of MEGS
Currently, there are no commercially-available power simulation platforms available that assess
the multi-scale, multi-system impact of distributed load management technologies. To address
these issues, we developed the Model for Electric Grid Strategies (MEGS) using a hybrid
modeling approach. Our approach numerically solves only the most salient power system
equations at appropriate levels of detail in conjunction with agent-based models of consumers
including non-linear human response algorithms. In this way, MEGS captures the coupled
contribution of distributed control, centralized control, and human response to the dynamic
power balance in the face of added variability from wind and solar generators. To simulate the
economic impact of high penetrations of intermittent generation, MEGS uses four components
illustrated in figure 4. The object-oriented systems implementation of these four components is
shown in figure 5.

System Dynamics Simulation of Power System Phenomena

Figure 6 illustrates the classical power system feedback loops present in MEGS. Component
behavior as outlined in figure 4 is color coded in figure 6. Renewable generation has a largely
exogenous effect on total generation and therefore is not included in the loop diagram. Figure 7
shows the implementation of the stock-and-flow structure of power grid physics within MEGS.
This implementation represents the blue variables in figure 6. Equations that describe both
physical and human behavior are implemented using the system dynamics methodology
described by Sterman [8] and Ford [9]. System dynamics is useful for computer simulation of
complex interactions between disparate variables with explicit implementation of feedback. The
ease with which disparate variables are incorporated allows MEGS to simulate, for example, the
impact of weather events on renewable generation or the impact of adjusted holiday schedules on
load profiles. This allows the modeler is able to construct a hypothesis about evolving causal
interactions through time using a visual programming language consisting of accumulators
(stocks), flows, and delay-inclusive feedback.

5
Figure 4. Top-level components of the MEGS modeling platform

Figure 5. Multi-object structure of the Model for Electric Grid Strategies (MEGS) showing
feedback between system physics, generation, and load. MEGS contains a visual object structure
that allows algorithms to be easily exchanged and feedback to be visualized.

6
Figure 6. Causal loop diagram of power grid physics (blue), generation controls (orange), and
load behavior (yellow).

Figure 7. The dynamic energy balance of a power system is constructed using the system
dynamics methodology. One feedback loop is apparent in this diagram, between angular
acceleration and rotational speed.

7
Agent-Based simulation of individual load behavior
To manage the complexity inherent in a multi-scale approach, MEGS utilizes an agent-based
modeling methodology. As described by d’Inverno and Luck [10], the agent-based methodology
offers an abstraction tool for systems with multiple interacting autonomous components. Using
MEGS, the aggregated contribution of thousands to millions of individual load management
devices can be accurately simulated in a fully customizable environment. Simulation
experiments may also be constructed regarding the individual control settings of distributed
devices. In MEGS, consumers and their load devices are the primary agents. Figure 9 illustrates
the graphical design of the residential consumer agent. The types of loads common to residential
consumers are explicitly simulated within this agent, such as electric water heaters for residential
consumers. The residential consumer has agency because she is able to make autonomous
decisions given data about the system. In most systems, these decisions slowly evolve based on
the limited amount of data utilized by consumers (e.g. end-of-month billing data). As
computerized autonomous load management technologies are added to the consumer’s
household, data availability and decision speed may be improved. This approach offers an
unprecedented description of load behavior in response to both internal and external drivers. By
using consumer agents, MEGS begins to examine the behaviors that cause load to exhibit
particular patterns instead of merely empirically including these patterns.

Figure 8. Agent-based construct for residential load behavior

MEGS simulates multi-timescale phenomena

With MEGS, feedback loops can be explicitly included at multiple levels of abstraction. For
example, figure 5 shows feedback loops at the coarsest level of granularity. The frequency of the
power system affects the response of both generation and load, which returns to affect the
frequency. At a finer level of granularity, figure 7 illustrates feedback between the rotational
speed of generators and their acceleration. By including only the causal relationships that drive
the behavior of interest and excluding extraneous detailed calculations, MEGS uses system
dynamics to achieve speed and clarity.
MEGS currently simulates the dynamic power balance of a utility grid using the system
dynamics methodology to numerically integrate the differential equations for system-wide

8
frequency and angular displacement. The core differential equation to this dynamic is the swing
equation, as described in detail by Kundur [11]. Because MEGS solves this equation along with
control system logic and agent-based contributions very quickly, it is able to simulate system
frequency at a 0.1 s resolution. Currently it is able to simulate a 24 hour period of frequency
dynamics in less than 1 minute. These fast run speeds are critical to being able to reflect
dynamics at multiple scales, because the finest scale will likely be the computationally-limiting
factor. In addition to the swing equation, the types of dynamic behaviors currently included in
MEGS are generator governor controls, centralized automatic generation controls, economic
dispatch, generator scheduling, stochastically-sampled wind and solar generation, stochastically-
sampled load, weather-driven load and consumer-driven load. MEGS achieves further
computational efficiency by calculating dynamic changes in these behaviors at the appropriate
time resolution. For example, diurnal dynamics are calculated at a 1 hour resolution and
interpolated every time step instead of being directly calculated at each time step.

Several assumptions have been made to allow MEGS to have fast run speeds. The first and most
scope-limiting assumption that MEGS makes is that the system is a very tight frequency island,
meaning that differences in speed from generator to generator are much smaller than the change
in system frequency due to exogenously-driven power imbalances. This also means that there is
no tie-line coordination with external power systems, and therefore MEGS currently reflects
dynamics on islanded systems only. The second major assumption of the current MEGS
platform is that voltage issues do not significantly affect frequency during normal operations.
Because MEGS is not meant to be a full electromagnetic transient simulator, this assumption has
validity. Nonetheless, during significant power swings, generator excitation and reactive power
flows can affect the flow of real power, and ultimately the system frequency.

Simulating contributions of intermittent generation

The accurate simulation of intermittent generation over multiple timescales will be critical to
simulating the overall impact of increased renewable energy penetrations. Currently, MEGS
simulates the contribution of power from wind plants and solar photovoltaic plants. A mean daily
profile for each type of intermittent generator is imported, and stochastically sampled noise is
added to this mean profile to simulate the aggregate unforecasted behavior. This noise may be
sampled over multiple periods. For instance, in the case of solar generation, several high-
frequency components of variability exist, and therefore short periods are used. In the case of
wind, longer periods are acceptable. Daily profiles of wind and solar generation as simulated in
MEGS are illustrated in figure 9.

9
Figure 9. Daily generation profiles for (a) solar and (b) wind as simulated by MEGS with
different components of stochasticity.

Simulating economics and real-time dispatch

To understand the difference between operational costs given multiple scenarios, in-depth
economics are integrated into MEGS. Currently, MEGS incorporates a simplified economic
dispatch algorithm into its calculation of dispatched generator unit commitments within the
dispatched generation component. This algorithm assumes that generators have a discrete and
constant cost relationship to one another, and therefore the economic priorities for operation are
unchanging. This will be insufficient for simulating realistic operation once systems become
large and generators exhibit non-linear cost behavior. To accurately simulate a dynamic
economic dispatch, mixed integer programming techniques could be employed that solve for unit
commitments given several constraints and objectives, such as cost minimization, reserve
margins, and ancillary service contributions. Again, the tradeoff between speed and realism is
weighed as additional calculations are incorporated.

Simulating load behavior with load management

The functionality of MEGS that has the highest impact on a prosumer paradigm shift is the
realistic and fully-coupled simulation of load in response to grid conditions. Currently, MEGS
simulates consumers as probabilistic agents, each having unique behavior in response to their
own preferences and system variables such as frequency. Using this framework, consumer agents
make autonomous decisions about when to increase or decrease load. In a scenario with no load
management, these decisions are not based on grid conditions. Figure 10 illustrates results of a
simulation in which consumers have no reactive response. Multiple agents combine to make the
overall load profile to the utility. Each agent has preferences about, for example, when to wash
dishes and take showers as illustrated by the timing of major household loads at the top of the
diagram. As these decisions are aggregated, a realistic load profile is generated.

10
Figure 10. Multiple consumer agents contribute to the overall load profile.

Model Validation and Preliminary Studies

The MEGS model was compared to and validated against data from an electric utility that
experienced a slow generation failure, triggering a frequency deviation event that lasted several
minutes. Approximately 140 MW of generation was gradually lost over a period of five minutes.
Available system data included generation, load, and frequency as functions of time. A
simplified MEGS simulation was created to validate the model by disabling or removing all the
control algorithms for generation and load management initially developed for MEGS. The
recorded data were used to calculate apparent total system inertia and then to “tune” the model to
the specific physical system properties by matching the general characteristics of the frequency
data. Figure 11 shows the results of the tuned model compared to the actual frequency recorded
during this event.

Hz

59.9

59.8

59.7

59.6
12:00 AM 12:01 AM 12:02 AM 12:03 AM 12:04 AM 12:05 AM

C alculate d Fre que ncy (Hz) Initial Fre que ncy (Hz) R e corde d Fre que ncy (Hz)

Figure 11. Frequency as a function of time for the simplified MEGS model compared to actual
data.

11
The inertia-tuned frequency response of figure 11 gave us confidence in the model’s ability to
represent system frequency given power imbalances. Once the concept of frequency response to
generation imbalance was studied, we tested an event on the larger Electric Reliability Council
of Texas (ERCOT) system with an approximate 8% loss in generation. ERCOT is one of the
Independent System Operators in North America Electric Reliability Corporation and provides
power to nearly the entire state of Texas. This frequency event is shown in figure 12a, adapted
from a study by Kirby et al. [12] which examined the relationship between frequency and power
markets. While the study concluded that frequency deviations - some of which are due to higher
renewable energy penetrations - do not pose a major reliability risk, they also show that the
power trading market and scheduling algorithms have sizable room for economic improvement.
After tuning the MEGS model using the generation control points (the dark orange squares in
figure 7), we obtained figure 12b for the frequency response. The drop in frequency is similar to
that for the observed ERCOT system, and the time to recover is approximately 9 minutes for the
observed and simulated frequencies. The simulation does not vary load, but only drops the
generation, which is likely the reason for our lack of sub-minute frequency variability. Also, the
simulated frequency response asymptotically approaches a higher-than-nominal 60.01 Hz. The
extra 0.01 Hz is seen because the simulated operators are trying to “make up” for the time spent
below nominal frequency from a total energy sales accounting viewpoint.
a) b)
Hz
60.05

60.025

60.00
Frequency

59.975

59.95

59.925

59.90
12:00 AM 12:05 AM 12:10 AM 12:15 AM 12:20 AM

Figure 12. a) Frequency as a function of time for an actual loss-of-generation event in the
ERCOT system [13]. b) Frequency simulated by MEGS to the same loss-of-generation event.

Once we developed confidence that MEGS models can simulate the frequency behavior of a
large system such as ERCOT, we tested the benefit of responding to frequency variations using
load management. With 300 MW of frequency-responsive load being controlled by a
proportional-integrative-derivative controller, we obtained the frequency response depicted in
figure 13. Notice that the frequency deviation is smaller but the recovery time is similar. The
over-frequency make-up response is also lower in magnitude.

12
 Frequency during loss of generation event with
and without responsive load
Hz
60.05

60.025

60.00

Frequency
59.975

59.95

59.925

59.90
12:00 AM 12:05 AM 12:10 AM 12:15 AM 12:20 AM

Figure 13. The response of the simulated ERCOT system to an 8% loss of generation is shown
without a load management scheme in grey hashing, and with a load management scheme in
solid black.

To investigate the improvement in economic operations and to begin to place a value on grid-
responsive load, we added further economic calculations based on generator operation in MEGS.
Shown in figure 14, the generators have an aggregated thermal efficiency curve, which controls
the amount of fuel that they collectively need for a certain generation target. Generators are
slightly more efficient at higher speeds, as indicated by the effect of speed on efficiency. Fuel
prices were assumed to be near nominal natural gas prices in 2008, namely 6.8E-9 dollars per
Joule [13]. Additionally, we assumed that generators operate less efficiently during ramping
conditions such as those likely experienced in the first few seconds of an event. This behavior is
reflected in the Gen ramp cost variable in figure 14, which itself is dependent on the rate of
change of generation, named the GEN SLEW RATE. This kind of behavior can be equated to the
efficiency lost during stop-and-go driving in a common automobile. We assumed ramping costs
$20 for every MW/s ramp. Based on these assumptions, using the simulation of an 8% loss of
generation on the ERCOT system, we simulated the operation cost during the event, which is
illustrated in figure 15b. The simulated frequency-responsive load is shown in figure 15a. With
only 300 MW of frequency-responsive load, which is approximately 1% of total load, the cost of
operation is over 2% lower. The reason is threefold. First, load responds more quickly than
generation to the event due to its low inertia and ability to employ derivative-based control.
Second, this fast response limits the need for generators to ramp quickly because load is now the
“first responder.” Third, generators can operate at a higher efficiency during the event because
the dropped load makes up nearly 10% of the initial generation imbalance and therefore the
speed deviation is smaller during the event.

13
 Gen ramp cost
per MW
GEN SLEW RATE GENERATION

effect of speed
Gen ramp cost
on efficiency
efficiency
Current change per unit
Thermal
efficiency speed change
efficiency curve
Speed
Fuel Price per
diversion
Joule

Fuel spend rate Operation Cost

dollar rate

Figure 14. Stock and flow diagram of power system operation costs.

a) b)
Operation costs with and without grid
Behavior of grid responsive load during responsive load
MW event USD/(MW*hr)
140
0

120
Operation cost per MWh

-100

100

-200
80

-300
60

-400
40
12:00 AM 12:05 AM 12:10 AM 12:15 AM 12:20 AM 12:00 AM 12:05 AM 12:10 AM 12:15 AM 12:20 AM

Figure 15. a) During the event, frequency-responsive load responds quickly and drops all of its capacity
until the frequency recovers. b) Operation costs are lower with the frequency-responsive load online and
remain lower for the entire simulation.

Conclusions

Application of system dynamics to study the control complexity of the United States electric
power grid shows great promise for modeling and understanding the effects of renewable energy
and smart-grid load management techniques and devices. We have constructed and
demonstrated a model of a simplistic islanded power system to show and analyze the effects
renewable generators including multiple levels control from the unit level to regional area
control level. The beginnings of advanced load management techniques at the residential level
have been introduced by treating households as agents with distinct personalities and properties
from a behavioral viewpoint. This is especially important because combining the methods of
agent-based and system dynamics modeling minimizes complexity, improves realism, and
increases speed for power system dynamic frequency simulations.

The physics of the methodology has been validated by reproducing and studying two separate
frequency excursions on real systems with the model encompassing the general behavior and
characteristics of each event. The MEGS model was successfully tuned to replicate a small
utility frequency excursion where 140 MW representing approximately 10% of the nominal

14
generation was lost over a three minute time period with the subsequent system recovery as more
generation combined with some load shed was used to stabilize the event.

On a larger system, the MEGS model was able to replicate the response characteristics of an
excursion on the ERCOT system. Based on the response to this event, the ERCOT system
appears good reliability characteristics and is capable of absorbing large amounts of variability
from renewable generators. The beginning of economic analysis was introduced into the model
to show how frequency excursions might impact the economic efficiency of the grid in the
ERCOT system. As indicated by the spike in operating costs (figure 14b) when large MW/s
ramping rates are required, increasing either the frequency or the magnitude of power variability
may greatly increase the long-term cost of operation. We have shown that frequency-responsive
load management can be nearly twice as effective at lowering the cost to respond to grid
frequency perturbations as current operations are now using common generators. Therefore, the
opportunity exists for utilities or power marketers to offer similar ancillary service incentives to
the owners of grid-responsive load devices as they now offer to owners of generation. A number
of new grid management strategies become available using these techniques. For example, even
with a very large drop in generation, frequency-responsive load devices that provide valuable
grid stabilization could return online within a few minutes of the event until generation stabilizes
and recovers. However, if necessary, the loss of these devices for short periods of time can also
be accounted and assessed for any loss of comfort or reliability to the smart device owner. These
types of analyses can be readily incorporated into the system dynamics methodology developed
and presented in this work.

15
References
1. Frequency Regulation for the Grid, A Dehamna, K. Adamson, Pike Research, October 2012.
2. Energy Storage Systems Cost Update, S Schoenung, Sandia National Laboratories,
SAND2011-2730, April, 2011.
3. George W. Bush Presidential Executive Order, January 24th , 2007.
4. Renewable Portfolio Standards in the States: Balancing Goals and Strategies, K. Cory and B
Swezey, National Renewable Energy Laboratory Technical Report, NREL/TP-670-41209,
December 2007.
5. Y. Wan, “Analysis of Wind Power Ramping Behavior in ERCOT”, National Renewable
Energy Laboratory Technical Report, NREL/TP-5500-49218, March 2011.
6. Nexight Group, Electric Power Industry Needs for Grid-Scale Storage Applications, Joint
Publication Sponsored by DOE Office of Electricity Delivery and Energy Reliability and
DOE.
7. Energy Storage – An Asset to Ontario’s Smart Grid, Presentation by the Corporate Partners
of the Smart Grid Forum, Energy Storage Working Group, April 2012.
8. Sterman, J.D., Business Dynamics: Systems Thinking and Modeling for a Complex World,
Irwin/McGraw-Hill, 2000.
9. Ford, A., Greening the Economy with New Markets, Proceedings of the 28th International
Conference of the System Dynamic Society, July 25-29, 2010, Seoul, Korea.
10. D’Inverno, M. and Luck, M., Understanding Agent Systems, Springer-Velag Berlin and
Heidelberg GmBh & Co. K, ISBN 3642073824, 2010.
11. Kundur, P., Power System Stability and Control, McGraw-Hill, ISBN 0-07-035958-X, 1994.
12. Kirby, B., Dyer, J., Martinez, C., Shoureshi, R., Dagle, R., Frequency Control Concerns In
The North American Electric Power System, ORNL/TM 2003/41, Oak Ridge National
Laboratory, Oak Ridge TN, December 2002.
13. Laughlin 2008, http://large.stanford.edu/publications/coal/references/gasprice.

16