This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.DOI

Behavioral Modeling for Microgrid

Simulation
THOMAS DEETER1 , DAISY H. GREEN 2 , (STUDENT MEMBER, IEEE), STEPHEN KIDWELL 3 ,
THOMAS J. KANE 3 , JOHN S. DONNAL 4 , (MEMBER, IEEE), KATHERINE VASQUEZ 1 ,
BARTHOLOMEW SIEVENPIPER 1 , AND STEVEN B. LEEB 2 , (FELLOW, IEEE)
1
U.S. Navy, USA
2
Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
3
U.S. Coast Guard, USA
4
Department of Weapons and Systems Engineering, U.S. Naval Academy, Annapolis, MD 21402, USA
Corresponding author: Daisy H. Green (e-mail: [email protected]).
This work was supported in part by the Office of Naval Research NEPTUNE Program, and in part by the Grainger Foundation.

ABSTRACT Trends in power system simulation that demand computationally-intensive, physics-based

models may impede the acquisition of useful results for applications like condition-based maintenance,
electrical plant load analysis (EPLA), and the scheduling and tasking of finite generation and distribution
resources. A tool that can quickly evaluate many scenarios, as opposed to intense, high fidelity modeling of
a single operating scenario, may best serve these applications. This paper presents a behavioral simulator
that can quickly emulate the operation of a relatively large collection of electrical loads, providing “what-if”
evaluations of various operating scenarios and conditions for more complete exploration of a design or plant
operating envelope. The presented simulator can provide time-series data of power system operation under
loading conditions and usage assumptions of interest. Comparisons to field data collected from a microgrid
on-board a 270-foot (82 meter) US Coast Guard medium-endurance cutter demonstrate the utility of this
tool and approach.

INDEX TERMS Load modeling, power system planning, power system simulation

I. EFFICIENT EMULATION systems can be physically-modeled based on thermodynamic

PPLYING the remarkable gift of vast computational principles [6]. Physics modeling is always an effort, and
A capability in an endless drive for increased simulation
fidelity with the physical world may offer diminishing returns
accurate parameter identification is essential [5], [6]. Alterna-
tively, with measurement-based modeling, in which data ac-
in many design and analysis efforts. For a power system, quisition equipment is used to measure actual load behaviors,
high fidelity simulations might be irreplaceable for fault accurate load characteristics can be obtained quickly from
studies, analysis of the effects of non-linear loads, or the a physical example, when available [7]. Curve-fitting tech-
subtle impacts of changing environmental conditions, e.g., in niques in sophisticated forms, e.g., artificial neural networks,
temperature or humidity, on electromechanical performance have been applied for measurement-based modeling and the
[1]–[3]. However, particularly for analyses focused on “what- efficient representation of complex nonlinear systems [8].
if” studies (which may explore many possible operating sce- However, it is challenging to obtain data over a wide range of
narios and conditions), current trends in power system sim- operating conditions for use in measurement-based modeling
ulation that demand computationally-intensive, physically- [5], [9]. There has also been interest in using residential
based models may impede the acquisition of useful results user behavior for predicting energy use patterns and demand
[4], [5]. [9]–[11]. In [12], both user behavior and individual load
There are two main approaches for load modeling component models are used together for residential power
in power system simulation: physically-based modeling system modeling.
and measurement-based modeling. Physically-based, or Power system design and analysis, especially for isolated,
component-based modeling, uses physics and mathematical microgrid, and generation-constrained systems, requires flex-
descriptions to characterize a load. For example, HVAC ible evaluation of “what-if” scenarios to design adequate

VOLUME 4, 2016 1

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

capability and to explore options for condition-based main- assumptions of interest. The resulting waveforms can be used
tenance [13], electrical plant load analysis (EPLA [14]), to evaluate typical likely demand profiles on the grid. The
and the scheduling and tasking of limited generation and SPS output waveforms can also be used to test monitoring
distribution resources [15]. “What-if” scenarios can reflect schemes for condition-based maintenance by including ex-
different stages of operation of loads, while also accounting emplar waveforms for pathological conditions inserted ran-
for load dependencies and the effects of exogenous variables domly or on other schedules into the simulation. Further-
such as time of day, weather, or human behavior [4]. For more, the SPS outputs can be used to confirm the starting
example, a “what-if” study on a ship might vary the number assumption that the grid will behave as a relatively stiff,
of propulsion shafts energized at a time. A tool that can decoupled collection of loads. Output waveforms that would
quickly evaluate many scenarios, as opposed to intense, high- challenge this assumption, i.e., overload the grid capability,
fidelity modeling of a single operating scenario, may best can be checked to confirm when the SPS assumptions are
serve these applications. Behavioral modeling of electrical trustworthy.
loads [16] may provide sufficient fidelity to provide datasets Behavioral emulation of loads in the SPS begins by de-
and enable rapid exploration of a range of scenarios. Be- scribing a load as a power consumer represented by a finite
havioral modeling combines rules about load behavior, time state machine (FSM) [17]. Temporal transitions in the FSM
series data, and stochastic modeling to develop realistic load power demand model can follow a fixed logical sequence
profiles and evaluate electrical demand within the system [4]. in time, or transition between states randomly, or in con-
This paper introduces the Shipboard Power Simulator cert with the states of other systems or parent systems on
(SPS), a software tool configured as a behavioral simula- the microgrid, or in response to environmental conditions
tor that can quickly emulate the operation of a relatively like temperature. Proper modeling of the ship (microgrid)
large collection of electrical loads on a microgrid, providing operational dependencies becomes more important than the
“what-if” evaluations to support more complete exploration precise mathematical description, e.g., differential or differ-
of a design space or plant operating envelope, and is further ence equations, for a load. Load power and load transients are
detailed in [17]. A wide variety of approaches have been modeled by stored waveform segments that represent power
adopted for time-domain simulation of microgrid and ship- demand in each state and in the transitions between states.
board power systems, as described in [18] and its associated The waveform segments can be acquired from field data if
references. The complexity of detailed time-domain simula- examples of the loads or microgrid system already exist, e.g.,
tion for power systems has pushed these simulation studies using a Nonintrusive Load Monitor (NILM) [13]. Alterna-
to hardware-in-the-loop and custom processors [18]. These tively, a waveform segment can be entered as a “best guess”
approaches are expensive, customized to particular problems from “name plate rating” information, or based on observa-
or power systems, and remain computationally burdensome tions of a similar load, or from a detailed load simulation
and time-consuming. In studies that require quick analysis conducted once, thus limiting the computational expense to
of many scenarios to understand energy consumption and a small set of expensive simulations with outputs that can be
aggregate dynamics, a behavioral simulation can provide re-used. This paper presents comparisons of SPS results with
speedy results without the computational burden of detailed field data collected from the power system on-board a 270-
time-domain or hardware-in-the-loop simulation. This paper foot (82 meter) US Coast Guard medium endurance cutter,
introduces a behavioral simulator unlike other time-domain SPENCER (WMEC 905). The SPS is evaluated against ship
simulation approaches for power systems. The SPS tool can data from SPENCER to demonstrate it’s ability to accurately
provide estimated load demand and EPLA load factors for a model ship behavior for various “what-if” scenarios, such as
load across multiple mission sets. By providing an accurate ship operational status, or “mission”, and a load operating in
picture of likely power demand, the data is invaluable when a degraded condition.
designing marine microgrids and assessing power system
operation. II. MARINE MICROGRIDS
In situations where the power system voltage is relatively The structure of the SPS behavioral modeling rules reflects
“stiff,” the interactions between loads on the system are the top-down design of a microgrid power system. An en-
relatively minimized. In this case, individual observations of gineering “V” [19] diagram illustrates the correspondence
loads can serve to represent the load behavior, reasonably between power system design and SPS modeling. The left
decoupled from the direct operation of other loads. The side of Fig. 1 visually illustrates the microgrid power sys-
SPS exploits this assumption to simulate the overall ship or tem design, beginning at the top with mission requirements
microgrid energy requirements as a function of different ship and operating environment, then engineering systems, and,
or grid missions. The intent of the SPS is not necessarily to finally, engineering electrical loads. Mission capabilities for
reproduce the exact instant-to-instant behavior of measured the microgrid, or ship in this case, drive the deployment of
data from an overall observation of a power system, although necessary ship systems. For example, requirements for speed
the simulator can be configured to do so. Rather, the goal and survivability may drive the number of propulsion shafts
of the SPS is to produce detailed time-series snapshots of installed on a ship. Each shaft requires necessary support-
power system operation under loading conditions and usage ing subsystems or electrical loads to provide compressed
2 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

FIGURE 1. Systems engineering “V” process reflected in Shipboard Power

Simulator (SPS) architecture.

FIGURE 3. Partial schematic of radial ac electrical distribution onboard

USCGC SPENCER, with details on the PORT and STBD subpanels.

FIGURE 2. Schematic of radial ac electrical distribution onboard USCGC

SPENCER.
ing systems, energized when the main engines are online,
include a variety of subsystems with pumps and other elec-
trical devices to support propulsion. Other example ship sys-
air, lubrication pumping, fuel pumping, and so on, each of tems include seawater cooling, electrical generation, HVAC,
which places demands on the ship electrical system. Each weapons systems, and communications.
supporting subsystem includes electrical loads like motors, Finally each ship system consists of various electrical
resistive heaters, and electronic controls. These loads operate loads. Because the electrical loads are part of ship systems,
at a specified voltage (e.g., 120V, 208V, or 440V ac) and their behavioral model can be determined by the operation
present two or more “states” of operation (e.g., “off” and of the system. Here, it is assumed that the ship microgrid
“on”), with specified power consumption levels. Connections is wired to provide radial ac electrical distribution. This is
may be line-to-line, three phase, and so forth. a common electrical distribution architecture used by the
High-level global parameters, such as the ship’s mission, USCG and USN. For example, Fig. 2 shows a depiction of the
the operating environment, or other factors further discussed electrical distribution of USCG cutter (USCGC) SPENCER.
in Section III of this paper, create requirements for specific A variety of sources can energize the 440 VAC, 60 Hz main
systems on ships, and cause these systems to energize or switchboard, depending on if the ship is at sea or in-port.
secure at any point in time. US Coast Guard (USCG) and US When at sea, power is provided by two ship-service diesel
Navy (USN) ships designate specific operational status, or generator (SSDG) sets, while in-port power is provided by ei-
“mission,” to indicate operating environment and ship system ther an aft or forward tie to shore power. These generator sets,
configuration. In the USCG, for example, “Alpha,” “Bravo,” along with the main propulsion diesel engines (MPDE) pro-
and “Charlie” mission statuses respectively indicate that the pelling the ship, are located in the ship’s engine room [20].
ship is underway, or in-port and ready to get underway if the The SSDG and MPDE require auxiliary equipment, e.g., fluid
need arises, or in-port and not expected to get underway for pumps and heaters, to maintain operational readiness when in
an extended period of time. Each mission calls for certain standby mode. Two electrical subpanels, port and starboard,
systems to be energized, and systems will be progressively which power these loads along with several other engine
brought from secured status into standby and, eventually, on- room loads critical for ship operation, are depicted in Fig.
line, or vice-versa, as the ship transitions between missions. 3.
Furthermore, environmental factors can also cause systems
to be energized or secured, such as HVAC loads during III. BEHAVIORAL MODELING
changing weather. The SPS is written in C++, and applies a “bottom-up” ap-
The engineering systems serve the different requirements proach [10], [21] for defining the hierarchical structure of a
of ship missions. For example, the shafting systems translate behavioral microgrid simulation, shown in the right side of
energy from the main engines into propulsion thrust. Shaft- Fig. 1. Beginning with individual loads, FSM power demand
VOLUME 4, 2016 3

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

erations Table. The “Loads Table” stores basic characteristics

for each load, including the load name, the voltage required
for operation, how the load is connected electrically (e.g, 3-
phase), as well as the FSM model of the load and its parent
system, if applicable. The “Power Trace Table” stores the
FIGURE 4. Finite State Machine (FSM) model of a CPP pump. power waveforms for steady-state and transient behavior at
a 10 Hz sample rate. The power waveforms can be from field
data, e.g. collected with a NILM, or they can be entered as a
graphical approximation of an expected power curve, or the
data can be provided from other sources including a detailed,
one-time simulation of the load. The “Uploaded Data Table”
serves as an index that identifies data in the the Power Trace
Table, e.g., as real or reactive power or harmonic content, and
with electrical phase for each waveform. The “Operations
Table” is another indexing table that connects a previously
designed FSM operation with its specific power trace.

B. SHIP DATABASE
Moving up a conceptual level in the simulator organization,
a “system” defines a collection of loads that work together
FIGURE 5. Power waveforms for the CPP pump.
to provide a service to the ship. For example, a “starboard
shafting system” consists of Number (NR) 1 CPP Pump, 1A
Lubrication Oil Service Pump (LOSP), and 1B LOSP. All
models for each load are stored in the SPS Load Database. three of these loads would energize in response to any mis-
Next, information about the radial panel wiring in the ship, sion that demanded propulsion from the starboard shaft, and
system response to global inputs, and the load-to-system therefore required the use of the starboard shafting system.
relationships are all stored in the Ship Database. Finally, One or more state-change variables define the operational
the operating profile and missions that the ship carries out behavior of a system. On a ship, sometimes operation is “au-
during the simulated time are stored in the Mission Profile tomatically” triggered depending on conditions on or around
Database. With these three databases properly configured, the the ship, and other times the operation occurs “manually,”
SPS assembles operating power waveforms for any collection e.g., based on an operator’s demand. Examples of state-
of ship missions of interest. change variables that might automatically trigger a system
to operate include simulated or operational variables like
A. LOAD DATABASE crew size, time of day, day of week, ambient temperature,
The FSM model for each load captures the power demand seawater temperature, mission, and ship speed. Changes to
behavior of the load. Each state in the model represents these parameters can prompt responses from the systems and
a unique steady-state power level. Each transition between associated loads based upon pre-defined rules of operation.
states is associated with a transient change in power demand. In contrast, manually activated systems may not operate on
For example, Fig. 4 shows a portion of the SPS Load Design a crisp schedule or in tight coordination with ship mission.
Wizard page displaying the FSM model of a Controllable These manually activated systems might, for example, in-
Pitch Propeller (CPP) pump. The real and reactive power clude pumps for graywater disposal on the ship. These pumps
waveforms for all three phases for this load, as observed on run essentially randomly distributed around other events, e.g.,
USCGC SPENCER, are shown in Fig. 5. The power traces crew needs. In the SPS tool, a stochastic state variable might
are segmented into three sections, the “Off-On” transient, command the operation of these “manual” loads.
“On” steady-state, and the “On-Off” transient. When first The Ship Database in the SPS stores the structure of the
activated, an in-rush current causes a large peak in power de- electrical panel system and load connections on the ship,
mand, creating the “Off-On” transient. Once the CPP pump and the relationship between each load on the power grid
finishes its transition to the “On” state, the “On”, or steady- and its associated system. The individual load behavior, e.g.,
state behavior, is repeated or concatenated in the behavioral the FSM describing the load’s power demands, have already
simulator until the load switches to the “Off” state. In steady been defined and stored in the Load Database. When a user
state, the pump is observed to consume approximately 2500 of SPS “connects” a load to the ship power system modeled
W and 1700 VAR per phase for real and reactive power, in a particular ship simulation, the Ship Database stores the
respectively. electrical interconnection or panel location of the load on
Critical information for the SPS regarding each load is the ship, and also stores the specific ship system associated
stored in a Load Database that comprises four tables: Loads with the load. For example, the NR 1 CPP pump might have
Table, Power Traces Table, Uploaded Data Table, and the Op- been defined previously in the Load Database as being part
4 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

of a “shafting system”. The Ship Database stores the data reflect the record of a real or imagined deployment of the ship
that pairs the NR 1 CPP pump with the specific shafting and its microgrid. For example, every ship maintains logs
system on the ship, in this case, “starboard shaft”. The that record the ship’s speed, ambient temperature, seawater
Ship Database also stores the associations or state-change temperature, crew size, date, present mission, engineering
variables that will trigger the operation of each system. Note plant line-up, and time. State-change variables affect the
that the Load Database defines load behavior specific to each behavior of the electrical loads. Data stored in the Mission
load, and the Ship Database stores load connections and Profile Database reflects the environment and events the ship
system memberships associated with a particular ship. This will experience during a simulation. Changing entries in
distinction permits a single Load Database, e.g., for a broad the log stored in the Mission Profile Database will create
class of USCG or USN surface ships, to serve as basic data endlessly varied simulations that assemble and emulate the
for assembling simulations of many different ships, each with behavior of collections of loads on the ship.
a distinct Ship Database, while reusing the same basic loads
where appropriate. IV. CONFIGURING AND RUNNING A SIMULATION
More specifically, the Ship Database comprises five tables: The user interface of the SPS eases the data input effort
Ship Electrical Network Table, Component Details Table, to quickly produce valuable results for “what-if” scenario
Panel Details Table, Load Interactions Table, and Mission analysis. This section demonstrates configuring the SPS to
Assumptions Table. The “Ship Electrical Network Table” simulate part of the USCGC SPENCER, with the radial
stores the structure of the ship’s electrical distribution system. electrical distribution system and loads shown in Fig. 3.
This table records the breakers, panels, transformers, and Configuring the SPS proceeds with the same “bottom-up”
interconnections that make up the electrical network. The approach used to describe the SPS database structure. First,
“Component Details Table” stores specifics on components the Loads Page allows the user to add, edit, delete, or preview
like panels in the network, including defining characteris- all loads within the Load Database. The Load Design Wizard
tics like voltages, connected phases, feeder conduit size, Page, shown in part in Fig. 4, is used to design or modify the
transformer type, transformer grounding, and transformer transient and steady-state behavior of a load.
sequencing for each component. The “Panel Details Table” Next, the Ship Design Page, shown in Fig. 6, configures
stores the specific details for a particular instance of a load the Ship Database. A user is given a “tree widget” to enter the
on the network and a tag indicating its defining example structure of the radial electrical network, from the generator
in the Load Database. For example, a load defined in the breakers to the power panels. Fig. 6 illustrates how the
Panel Details Table as “NR 1 CPP Pump” may be a specific user has designed the electrical network for the USCGC
instance of a load class “CPP Pump” found in the Load SPENCER, with two monitored panels connected to the Main
Database. The specific instance “NR 1 CPP Pump” is now an Switchboard. Specific loads, shown in Fig. 3, are entered as
example of the class with a particular connection to specific connections to specific panels. The user defines operations
power system phases, a defined location on the grid, and a for the ship-specific state-change variables using the Load
membership in a parent system like “starboard shafting”. The Class Interaction Designer pictured in the lower left corner
“Load Interactions Table” records the system membership for of Fig. 6. For example, a particular mission type known
each particular load, e.g., NR 1 CPP Pump to the “starboard as “Alpha Restricted Maneuvering Doctrine” (Alpha-RMD)
shaft system”. This table stores the specific state variables places a premium on the ships ability to move nimbly. In
and associated changes that influence the operations of each this case, both NR 1 and NR 2 CPP pumps are operated to
load. For example, if a load changes state with the ship’s provide the greatest likelihood of retaining propeller pitch
missions, the Load Interactions Table would have a record control, and these pumps change state when entering or leav-
of what transient operations a load would experience when ing the Alpha-RMD mission. When Alpha-RMD is entered,
a particular mission starts and ends. State change variables both CPP pumps energize, and when Alpha-RMD ends,
may be deterministic with mission changes, or stochastic both CPP pumps secure. The user defines the engineering
as a function of time or crew size or other environmental configurations and speeds at which this ship conducts various
variables. The SPS can model a load as responding to a missions, using the Mission Assumptions table.
logical combination of state change variables. For example, To load the Mission Profile Database, the user enters
a heater might respond to the logical AND of a deterministic mission logs on the Simulation Inputs Page. The user may
variable indicating that a system is in standby and a stochastic choose to manually enter the input variables such as mission
variable modeling thermostatic cycling. Finally, the “Mission and ambient temperature, or may upload a Mission Profile
Assumption Table” records ship operating rules including, CSV data file produced elsewhere, e.g., in a spreadsheet tool.
for example, the speed, generator configuration, and engine Fig. 7 shows a cropped image of the Simulation Inputs Page
configuration for different missions. after the user has uploaded a Mission Profile data file for a
15-day cruise.
C. MISSION PROFILE DATABASE Finally, the Reports Page allows the user to choose the
The highest level of data abstraction in the SPS, the Mission loads, phases and types of power to be simulated. Any panel
Profile Database, records a log of state-change variables that or breaker downstream of the selected breaker, or panel,
VOLUME 4, 2016 5

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

FIGURE 6. Ship Design Page.

of online, standby, and secured main engines and shafts for

each mission. The algorithm also logs the speed for each
mission recorded in the Ship Database. Once all the missions
have been processed, the SPS algorithm searches through the
other inputs in the Mission Profile Database such as seawater
temperature or crew size, and logs the start and stop time for
any changes. These changes define instants when a load may
change state.
Next, the user selects a collection of loads of interest. The
Ship Database is queried to identify phases associated with
each load. For each phase that supplies power to the load,
the SPS queries the Ship Database for the load’s state-change
variables and searches for changes in these variables in the
the Mission Profile Database. For all state changes from the
Mission Profile Database, the program uses the start and stop
times associated with the change to develop a list of possible
state changes for the load. Each of these times are checked
against the user defined state change variables in the Ship
FIGURE 7. Simulation Inputs Page. Database for the load. The start and stop times are computed
for the load based on the state changes, which includes both
the deterministic changes, such as mission inputs, as well as
is automatically included in the simulation. The user may stochastic changes, e.g., time, or crew size. The state changes
further elect to export the results to a text file. could also be a logical AND of a deterministic and stochastic
Fig. 8 illustrates the SPS algorithm to collect data from variable. For each state change, the SPS retrieves the name of
the Mission Profile Database, identify the operation of rel- the operation, such as “Off-On”, from the Ship Database.
evant ship systems from the Ship Database, and assemble For each state change, the SPS takes the time and operation
the power waveforms for relevant operating loads from the name, looks up the power waveform in the Load Database,
Load Database to produce simulation waveforms for the and adds it to the total time series for the phases connected to
emulated cruise. Simulation begins by searching through the load. In between state-changes, such as after an “Off-On”,
missions within the Mission Profile Database to create a set the SPS will retrieve the steady state, or “On” state waveform
6 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

“mission,” as the mission determines whether certain main

engines are online, in standby, or secured. The MPDE lube-
oil (LO) heater works in tandem with the pre-lube pump to
maintain desired lube-oil viscosity and temperature. These
heaters secure upon engine start, and return to an automatic
cycling mode when the engine secures. To model the heater
behavior in the simulator, a logical combination of two state-
change variables is used to trigger a change in the simulated
heater load state. A “primary” state-change variable tracks
with the the pre-lube pump and is activated whenever the
associated MPDE is put in standby. The “secondary” state-
change variable is a “random" event that models the au-
tomatic thermal cycling when the heater is operating. The
simulator effectively models the heater operation with a log-
ical AND operation of these two state-change variables, the
primary variable tracking the MDPE state and the secondary
variable tracking the random cycling. The heater energizes
and secures at random times to reproduce the thermostatic
behavior of the actual heater whenever the simulated MDPE
enters a standby state. Finally, the MPDE jacket water (JW)
heater consists of two 3-phase immersion heating elements
sitting between the cylinders on either side of the engine
FIGURE 8. Simulation Algorithm Process. block. These elements are modeled with the same approach
used for the LO heater.

from the Load Database and will repeat the waveform until B. SHIP SERVICE DIESEL GENERATOR LOAD
the next state-change, in this case the “On-Off” state change. BEHAVIOR
This process repeats until all requested phases, loads, and The Ship Service Diesel Generators (SSDG) are also served
power types have been analyzed across the entire mission by a collection of support loads. Similar to the MPDE LO
profile. The aggregate power is the summation of the power heater, the SSDG LO heater keeps engine lubricating oil
trace of each individual load. The next section demonstrates temperature within a set range while the engine is secured.
the utility of this approach by comparing SPS simulations to The SSDG LO heater is a line-to-line single-phase load. The
underway data recorded on a cruise of USCGC SPENCER. heater secures when the SSDG is brought online and the
temperature rapidly increases above the upper set point. The
V. COMPARISON TO FIELD DATA ship mission also drives the generator configuration, similar
Nonintrusive Load Monitor (NILM) systems are installed to the MPDE behavior. The primary state-change variable
upstream of two 440V sub-panels in the main engine room for this load is “mission” and the load will only activate if
of USCGC SPENCER. These monitors have proven to be the associated generator is put in standby. The secondary
reliable tools for collecting data for condition-based mainte- or “random” state-change variable models the thermostatic
nance, fault detection, and energy scorekeeping [20], [22]. cycling. The SSDG JW heater serves the same purpose as the
This NILM data provides insight into the load traces and MPDE JW heater and was modeled similarly.
behavior of equipment essential to the cutter’s operational
capability while underway. Fig. 3 lists the electrical loads C. ADDITIONAL ENGINE ROOM LOAD BEHAVIOR
on each of these two panels, port and starboard. This sec- The Controllable Pitch Propeller (CPP) pump is energized
tion discusses the observed load behaviors onboard USCGC when Alpha-RMD has been set. The primary state-change
SPENCER and describes how such behaviors were modeled variable is “mission”. The two graywater pumps are iden-
in the SPS tool, allowing for a comparison of the real ship tical, motor-driven centrifugal pumps that alternate runs to
data with the SPS simulation output data. empty the graywater tanks when a high level is reached. The
graywater pumps’ behavior was modeled through the use of
A. MAIN PROPULSION DIESEL ENGINE LOAD the random option for the state-change variable. The fuel
BEHAVIOR oil purifier (FOP) is run underway to clean the diesel oil
A collection of loads service each of the main propulsion before use in the MPDE or in the SSDG. Thus, it runs more
diesel engines (port and starboard). The main propulsion frequently if the quality of fuel onboard is poor. Because the
diesel engine (MPDE) pre-lube pump is a motor-driven cen- FOP is the only modeled load that is dependent on the fuel
trifugal pump that operates upon shutdown of the main diesel quality, this behavior was modeled using the “random” state-
engines. The state-change variable for the pre-lube pump is change variable. The in-port auxiliary saltwater (ASW) pump
VOLUME 4, 2016 7

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

behavior well.
Further, to test the SPS’s ability to accurately model
ship behavior during various “what-if” scenarios, the SPS
was used to simulate different ship missions. The simulated
power streams are used to analyze individual load operation,
by employing signal processing and pattern recognition algo-
rithms to associate the power waveforms with the operating
schedule of specific loads. The NILM algorithms can identify
load operation in both field and simulated data because the
SPS uses real “slices” of power observations to assemble
(a) SPENCER power streams as observed with NILM.
simulated waveforms. Loads are identified by first using a
change of mean detector to detect “on” and “off” events
[25]. The power stream can then be examined to calculate
a set of features. For the simulation data, the “on” and “off”
events are characterized using a correlation algorithm. The
correlation algorithm matches the shape of the input data to
known exemplars [26]. Consider two sampled waveforms f
and g, where f is an observation or input signal and g is a
load exemplar or example waveform. The correlation metric,
C, is
(f − f¯) · (g − ḡ)
(b) Simulation power streams from the SPS. C =1− 2 (1)
|g − ḡ|
FIGURE 9. 41-day aggregate power streams for the port sub-panel.
where f¯ and ḡ are the mean of f and g, respectively, and
are subtracted from the original signals, f and g in order to
provides cooling water at the pier. This pump was modeled remove the DC offsets. When C approaches zero, this indi-
with “mission” as its state-change variable, as it only turns cates that the exemplar and observation match in both shape
on when in port, and secures upon leaving port. The bilge and amplitude. When an event is detected, C is calculated for
and ballast pump is a motor-driven centrifugal pump used for each load using a fingerprint exemplar, and the event is clas-
taking on or discharging ballast water for stability purposes sified as the load with the minimum C value. For SPENCER,
and also for emptying the cutter bilges of excess water. The the loads are classified using the load identification algorithm
bilge and ballast pump was modeled as a two-state, random in [25].
occurrence.
A. CHARLIE STATUS
VI. SIMULATION RESULTS AND TESTING The first ship mission simulated was Charlie status, in which
The Shipboard Power Simulator was put through an exten- the cutter is moored at a pier and not expected to get under-
sive testing process to evaluate the simulator’s ability to way for an extended period of time. Results are presented for
accurately model ship behavior and estimated load demand. a six hour window of ship operations, with plots in Fig. 10
Unlike detailed time-domain simulation approaches which showing three-phase aggregate power time-series streams as
are often on the seconds timescale and require hardware-in- well as a NILM Dashboard “Timeline View” of individual
the-loop and computationally burdensome algorithms [18], loads [13]. The dashboard shows individual load operation
[23], [24], the SPS provides quick analysis of energy con- with colored bars to indicate periods a load is energized.
sumption and overall demand profiles for entire ship missions The six hour SPS emulation of Charlie status for the eight
and cruises. First, the simulator was challenged to reproduce active ship loads in this mission required only 35 seconds to
various ship missions, with testing culminating in a 41- complete on a 1.8 Ghz i7-8565 CPU.
day simulation shown in Fig. 9b, which spanned multiple The duty cycle, D, for each load is expressed as,
missions, speeds, and temperatures. This is compared to 41-
days of an observed cruise on USCGC SPENCER, from TON
, D= (2)
December 28, 2016 to February 7, 2017, shown in Fig. 9a. T
The purpose of the behavioral simulation, of course, is not to where TON is the total time a load is energized during
reproduce exact instant-to-instant details of the ships power time period T . Observed duty cycles from two observed
demands. Rather, the simulator reproduces the overall quality ship periods (December 29, 2016 and July 12, 2019) and
of the ship energy requirements as a function of different simulated duty cycles for the respective loads are compared
mission tasking. Comparing the observed power in Fig. 9a in Table 1. Fig. 10 shows the 2016 observed power streams
and the simulation output in Fig. 9b, both show similar and simulated power streams for the port and starboard power
aggregate power levels, indicating the SPS was able to model panels as well as the detected events during this 6-hour
8 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

(a) SPENCER power streams and detected events as observed with NILM. (b) Simulation power streams and detected events from the SPS.
FIGURE 10. Charlie status. Power streams are displayed above. NILM Dashboard Timeline View is displayed below, where colored blocks indicate periods where a
load is energized.

TABLE 1. Charlie Duty Cycle Results connection, but a minimum of one SSDG must be kept in
standby. As observed on SPENCER, and modeled in the
NILM NILM
Equipment
(2016) (2019)
SPS simulator, both SSDGs are in standby, with support heaters
Additional engine room loads for both generators cycling. The observed NR 1 and NR 2
Graywater Pump 0.003 0.007 0.009 SSDG LO heaters exhibited different behavior, due to the NR
In-Port ASW Pump 1.0 1.0 0.999 2 SSDG LO heater on the ship operating in degraded condi-
MPDE keep-warm system
NR 2 JW Heater 1.0 0.989 0.999 tion. This was modeled in the simulator, showing its ability
NR 1 JW Heater 1.0 0 0.999 to model this “what-if” scenario, in which both SSDG LO
SSDG keep-warm system heaters correctly respond to the primary state change variable
NR 2 JW Heater 0.417 0.424 0.483
NR 1 JW Heater 0.387 0.580 0.459
“mission” and activate when the associated generator is but in
NR 2 LO Heater 1.0 1.0 0.999 standby, but have different behavior based on the secondary
NR 1 LO Heater 0.556 0.377 0.481 state change variable representing thermostatic cycling.

B. ALPHA-NORMAL STATUS
Charlie status. The duty cycles for all loads agree well for the The next ship mission simulated was Alpha-Normal status, in
simulated data and both data observations, with the exception which the cutter is underway in open waters. Observed duty
of the NR 1 MPDE JW heater. The NR 1 MPDE JW heater is cycles from two observed 1-day ship periods (January 26-
not operating in the 2019 observation window due to a fault 27, 2017 and September 18-19, 2019), and simulated duty
condition in which the heaters were operating in a degraded cycles for the respective loads are compared in Table 2. Fig.
state with reduced power consumption or not operating at all 11 shows the 2017 observed power streams and simulated
[27]. The MPDE JW heaters were kept in healthy condition in power streams for the port and starboard power panels as
the simulation. For both the observed data and the simulation well as the detected load events during this 1-day Alpha-
data the main propulsion diesel engines are secured and kept Normal status. The SPS completed this simulation for nine
warm using only the jacket water heaters. The graywater active ship loads in 17 seconds on the i7 CPU. Again, the
is cycling, and the in-port ASW pump is on, as expected duty cycles for all loads agree well for the simulated data
while in-port. The ship is receiving power from a shore tie and both data observations, except for the NR 2 MPDE
VOLUME 4, 2016 9

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

(a) SPENCER power streams and detected events as observed with NILM. (b) Simulation power streams and detected events from the SPS.
FIGURE 11. Alpha-Normal status. Power streams are displayed above. NILM Dashboard Timeline View is displayed below, where colored blocks indicate periods
where a load is energized.

JW heater which is not operating in the 2019 observation TABLE 2. Alpha-Normal Duty Cycle Results
window due to the same MPDE JW heater fault condition
NILM NILM
observed in Charlie status. In this operational status, once the Equipment
(2017) (2019)
SPS
cutter leaves restricted waters and enters the open ocean, the Additional engine room loads
ship is configured for the Alpha-Normal condition and the Bilge and Ballast 0.025 0 0.002
CPP pumps are secured. One engine and one generator are FOP Centrifuge 0.055 0.024 0.055
FOP Feed Pump 0.060 0.029 0.063
secured and placed in standby status to save fuel, indicated Graywater Pumps 0.020 0.013 0.011
by the associated heaters and pumps cycling. In this case, MPDE keep-warm system
NR 2 MPDE and NR 1 SSDG are placed in standby in both NR 2 JW Heater 0.526 0 0.491
NR 2 LO Heater 0.796 0.734 0.885
the observed and simulated data. Additionally, the fuel oil NR 2 Prelube Pump 0.878 0.814 0.993
purifier may be online to move fuel to appropriate tanks SSDG keep-warm system
and the bilge and ballast pump may be online while moving NR 1 JW Heater 0.150 0.209 0.180
ballast water. Similar to Charlie status, the graywater pump NR 1 LO Heater 0.042 0.092 0.099
is seen cycling.
where Prated is the rated power, Pavg is the long term average
C. EPLA VALIDATION power the component will draw when in operation, T is the
Electrical Plant Load Analysis (EPLA) load factors are often total time period of interest, and P (t) is the power draw
called for in the design of marine microgrids. These load of the load at time, t. For the two-state loads common on
factors summarize power demand by providing an estimate the example power panel, P (t) is constant (Pss ) during the
of average load [4], [14]. The SPS tool can provide a wealth “on” state, as in Fig. 5. For these two-state loads, the average
of data for calculating EPLA load factors for a load across power can be expressed as,
multiple mission sets, providing an accurate picture of likely PN
(TON,i × Pss,i )
power demand. The load factor [4] is expressed as: Pavg = i=1 , (4)
T
Z T where i is an individual load cycle. There are N full or
Pavg 1
LF = , Pavg = P (t)dt, (3) partial load cycles in the time window, in which a full load
Prated T 0
10 VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

TABLE 3. EPLA Load Factor Results TABLE 4. Average Power

In-port Underway NILM Simulation Percent

NILM SPS NILM SPS Mission
Avg. Power Avg. Power Difference
Additional engine room loads Charlie 29.82 kVA 32.01 kVA 7.33%
Bilge & Ballast Pump 0 0 0.008 0.001 Alpha-Normal 19.42 kVA 19.45 kVA 0.16%
NR 2 CPP “C” Pump 0.099 0.121 0.064 0.064 In-port 33.66 kVA 35.85 kVA 6.81 %
NR 1 CPP “C” Pump 0.094 0.125 0.066 0.064 Underway 21.48 kVA 20.83 kVA -3.02 %
FOP Centrifuge 0 0 0.031 0.053
FOP Feed Pump 0 0 0.019 0.028
Graywater Pumps 0.015 0.011 0.020 0.016
In-Port ASW Pump 0.472 0.540 0 0 Charlie, Alpha-Normal, in-port, and underway), as well as
Main propulsion diesel engine (MPDE) keep-warm system
NR 2 JW Heater 0.958 1.01 0.376 0.344 the percentage difference. The 2016 and 2017 data from
NR 1 JW Heater 0.950 1.01 0.268 0.247 Table 1 and Table 2 is used for the Charlie and Alpha-Normal
NR 2 LO Heater 0.099 0.106 0.594 0.609 NILM observations, respectively.
NR 1 LO Heater 0.093 0.108 0.294 0.247
NR 2 Prelube Pump 0.093 0.111 0.614 0.688
NR 1 Prelube Pump 0.098 0.095 0.292 0.305 VII. CONCLUSIONS
Ship service diesel generator (SSDG) keep-warm system
NR 2 JW Heater 0.391 0.468 0.017 0.010 The duty cycle, EPLA load factors, and average power re-
NR 1 JW Heater 0.480 0.436 0.147 0.124 sults computed from the SPS emulations provide fantastic
NR 2 LO Heater 0.903 0.934 0.208 0.285 estimates in comparison to actual observations on-board a
NR 1 LO Heater 0.462 0.434 0.039 0.066
ship. The data provided by the SPS for the demonstrations
presented in the tables was available in minutes or less.
Of course, simulation through behavioral modeling is not
cycle consists of an “on” followed by an “off,” and a partial
expected to slavishly reproduce field waveforms. For exam-
load cycle is when either the detected “on” or “off” (or
ple, the NILM Dashboard plots of observed and simulated
both) of the full load cycle is outside of the time window of
Charlie status show clear differences between the exact time
interest. TON,i is the duration of each individual load cycle
instants of operation for the SSDG JW heaters. However,
(or portion thereof that is within t = 0...T ), and Pss,i is
they do reflect similar duty cycles of the field systems using
the steady-state power draw for each individual load cycle.
only seconds of computation to emulate a six hour operating
The variable Pss is calculated as the change in total apparent
window. Longer, multi-day missions were also emulated,
power, defined as the difference between the median values
providing accurate power demand estimates compared to
over ∆tM length windows, where here ∆tM = 0.5 seconds,
field data. The behavioral approach offers spectacular pos-
before and after the “on” to “off” event. Table 3 lists the
sibilities for quickly evaluating the energy demands and
load factor for each load, comparing the simulated data with
overall performance of microgrid loads over many operating
NILM data from USCGC SPENCER for two periods, in-
scenarios. Furthermore, the output of the simulator can be
port and underway, compiled from data from Fig. 9a. For
provided as base data for other studies, such as focusing
in-port data, several standard configurations during Charlie
design choices for electric distribution systems or sensitivity
and Bravo status from SPENCER were compiled into a 1-day
analysis to determine the reliability of metrics such as EPLA
in-port period. For underway data, several standard Alpha-
load factors [4]. Additional model input parameters can be
RMD and Alpha-Normal missions from SPENCER were
utilized for further studies. For example, using the quality of
compiled to create a standard 3-day underway period. The
fuel as an input parameter would enable studies of the effect
SPS required approximately 64 seconds and 151 seconds of
of fuel-quality on certain load profiles, such as the FOP. The
simulation time on the i7 computer for seventeen loads over
focus of this paper was a USCG medium endurance cutter,
the full emulated 1-day in-port period and 3-day underway
but this modeling approach can be applied to other small
period, respectively.
power systems or power systems that can be “islanded” in
general, such as a system with substantial renewable energy
D. AVERAGE POWER VALIDATION
and distributed resources. In this regard, future testing of the
The duty cycle results and EPLA load results show that SPS simulator includes simulating other types of microgrids
the simulator can accurately model individual load behavior. and distribution systems, such as a ring bus zonal electrical
Although the SPS is not meant to give an instant-to-instant distribution system, often used by the US Navy, or providing
replication of the observed data, it does give an accurate energy demand scenarios to aid in the facilitation of renew-
overall picture of power demand and load profiles. Thus, the able energy sources to a grid.
average power is calculated for each of the above simulated
missions to compare the actual observed ship data with the
SPS simulated output. The average power for each load from ACKNOWLEDGMENT
Eq. (4) is summed, to give the overall power demand from The authors gratefully acknowledge the support and dedica-
both the port and starboard subpanels. Table 4 shows the tion of the US Coast Guard, and, in particular, the spectacular
average power for the previously described missions (i.e., crew of USCGC SPENCER.
VOLUME 4, 2016 11

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

REFERENCES systems with distributed energy storage,” Journal of Marine Engineering

[1] J. V. Milanovic, K. Yamashita, S. Martínez Villanueva, S. . Djokic, and & Technology, vol. 19, pp. 31–44, 2020.
L. M. Korunović, “International industry practice on power system load [24] D. Kaza, C. Soldano, F. Silvestro, G. Schiapparelli, F. D’Agostino, and
modeling,” IEEE Transactions on Power Systems, vol. 28, no. 3, pp. 3038– M. Martelli, “Development of a multiphysics real-time simulator for
3046, 2013. model-based design of a dc shipboard microgrid,” Energies, vol. 13, July
[2] S. Shao, M. Pipattanasomporn, and S. Rahman, “Development of physical- 2020.
based demand response-enabled residential load models,” IEEE Transac- [25] D. H. Green, S. R. Shaw, P. Lindahl, T. J. Kane, J. S. Donnal, and S. B.
tions on Power Systems, vol. 28, no. 2, pp. 607–614, 2013. Leeb, “A multiscale framework for nonintrusive load identification,” IEEE
Trans. on Ind. Informataics, vol. 16, no. 2, pp. 992–1002, 2020.
[3] T. I. Bø, A. R. Dahl, T. A. Johansen, E. Mathiesen, M. R. Miyazaki,
[26] J. Paris, J. S. Donnal, and S. B. Leeb, “NilmDB: The non-intrusive load
E. Pedersen, R. Skjetne, A. J. Sørensen, L. Thorat, and K. K. Yum, “Marine
monitor database,” IEEE Transactions on Smart Grid, vol. 5, no. 5, pp.
vessel and power plant system simulator,” IEEE Access, vol. 3, pp. 2065–
2459–2467, 2014.
2079, 2015.
[27] D. Green, P. Lindahl, S. Leeb, T. Kane, S. Kidwell, and J. Donnal, “Dash-
[4] U. Orji, B. Sievenpiper, K. Gerhard, S. B. Leeb, N. Doerry, J. L. Kirtley,
board: Nonintrusive electromechanical fault detection and diagnostics,” in
and T. McCoy, “Load modeling for power system requirement and capa-
2019 IEEE AUTOTESTCON, Aug 2019.
bility assessment,” IEEE Transactions on Power Systems, vol. 30, no. 3,
pp. 1415–1423, May 2015.
[5] A. Arif, Z. Wang, J. Wang, B. Mather, H. Bashualdo, and D. Zhao, “Load
modeling - a review,” IEEE Transactions on Smart Grid, vol. 9, no. 6, pp. THOMAS DEETER received the Naval Engineer
5986–5999, 2018. degree and a Master’s degree in Engineering and
[6] F. Scotton, L. Huang, S. A. Ahmadi, and B. Wahlberg, “Physics-based Management from MIT in 2020. He currently
modeling and identification for hvac systems?” in 2013 European Control serves as an officer and active duty engineer in the
Conference (ECC), 2013, pp. 1404–1409. United States Navy.
[7] B. Choi and H. Chiang, “Multiple solutions and plateau phenomenon
in measurement-based load model development: Issues and suggestions,”
IEEE Transactions on Power Systems, vol. 24, no. 2, pp. 824–831, 2009.
[8] Q. Ai, D. Gu, and C. Chen, “New load modeling approaches based on field
tests for fast transient stability calculations,” IEEE Transactions on Power
Systems, vol. 21, no. 4, pp. 1864–1873, 2006.
[9] L. Bottaccioli, S. Di Cataldo, A. Acquaviva, and E. Patti, “Realistic multi-
scale modeling of household electricity behaviors,” IEEE Access, vol. 7, DAISY H. GREEN is currently pursuing a Ph.D.
pp. 2467–2489, 2019. degree in electrical engineering with the Research
[10] A. Capasso, W. Grattieri, R. Lamedica, and A. Prudenzi, “A bottom- Laboratory of Electronics, Massachusetts Institute
up approach to residential load modeling,” IEEE Transactions on Power of Technology. She received a B.S. degree from
Systems, vol. 9, no. 2, pp. 957–964, 1994. the University of Hawaii at Manoa in 2015 and an
[11] J. Widén and E. Wäckelgård, “A high-resolution stochastic model of do- M.S. degree from Massachusetts Institute of Tech-
mestic activity patterns and electricity demand,” Applied Energy, vol. 87, nology in 2018, both in electrical engineering.
no. 6, pp. 1880 – 1892, 2010.
[12] A. J. Collin, G. Tsagarakis, A. E. Kiprakis, and S. McLaughlin, “Devel-
opment of low-voltage load models for the residential load sector,” IEEE
Transactions on Power Systems, vol. 29, no. 5, pp. 2180–2188, 2014.
[13] A. Aboulian, D. Green, J. Switzer, T. Kane, G. Bredariol, J. Donnal, STEPHEN KIDWELL is a Lieutenant in the U.S.
P. Lindahl, and S. Leeb, “Nilm dashboard: A power system monitor for Coast Guard. He received a M.S. in Naval Ar-
electromechanical equipment diagnostics,” IEEE Transactions on Indus-
chitecture and Marine Engineering at the Mas-
trial Informatics, vol. 15, no. 3, pp. 1405–1414, Jun 2018.
sachusetts Institute of Technology in 2020. He
[14] N. Doerry, “Electric Power Load Analysis,” Naval Engineers Journal, vol.
served as a Student Engineer and eventually Assis-
124, no. 4, pp. 45–48, Dec. 2012.
tant Engineer Officer onboard USCGC HAMIL-
[15] H. Chin, C. Su, and C. Liao, “Estimating power pump loads and sizing
generators for ship electrical load analysis,” IEEE Transactions on Indus- TON (WMSL 753) from 2015-2018, and now
try Applications, vol. 52, no. 6, pp. 4619–4627, 2016. serves as an Availability Project Manager for the
[16] S. B. Leeb, J. L. Kirtley Jr., D. S. Woodruff, and M. Le Van, “Building- Coast Guard’s Icebreaking, Buoy, and Construc-
Level Power Network Analysis,” vol. 5, no. 1, pp. 30–35, Jan 1992. tion Tender fleet.
[17] T. Deeter, “Creating a Shipboard Power Simulation Tool Using Electrical
Load Behavior Modeling,” Master’s thesis, Massachusetts Institute of
Technology, 2020. THOMAS J. KANE is a Lieutenant in the U.S.
[18] K. Schoder, H. Ravindra, M. Stanovich, J. Langston, I. Leonard, and Coast Guard, stationed as the Engineering Officer
M. Steurer, “Shipboard power system baseline modeling and evaluation,” onboard USCGC CAMPBELL. He received an
in 2019 IEEE Electric Ship Technologies Symposium (ESTS), 2019, pp. M.S. degree in mechanical engineering at the Mas-
536–541. sachusetts Institute of Technology in 2019. He was
[19] “Systems Engineering Handbook: A Guide for System Life Cycle Pro- previously stationed as Damage Control Assistant
cesses and Activities, version 3.2.2N,” International Council on Systems aboard USCGC MELLON and as a Port Engineer
Engineering (INCOSE), San Diego, CA, USA. for the National Security Cutter (NSC) fleet.
[20] P. A. Lindahl, D. H. Green, G. Bredariol, A. Aboulian, J. S. Donnal,
and S. B. Leeb, “Shipboard fault detection through nonintrusive load
monitoring: A case study,” IEEE Sensors Journal, vol. 18, no. 21, pp.
8986–8995, Nov 2018.
[21] L. Chuan and A. Ukil, “Modeling and validation of electrical load pro- JOHN S. DONNAL is currently a Faculty Mem-
filing in residential buildings in singapore,” IEEE Transactions on Power ber of the U.S. Naval Academy, Annapolis, Mary-
Systems, vol. 30, no. 5, pp. 2800–2809, 2015. land, in Weapons and Systems Engineering. He re-
[22] T. Kane, “The NILM Dashboard: Shipboard Automatic Watchstanding ceived the B.S. degree from Princeton University
and Real-Time Fault Detection using Non-intrusive Load Monitoring,” in 2007 and the M.S. and Ph.D. degrees from the
Master’s thesis, Massachusetts Institute of Technology, 2019. Massachusetts Institute of Technology in 2013 and
[23] C. S. Edrington, G. Ozkan, B. Papari, D. E. Gonsoulin, D. Perkins, 2016, respectively, all in electrical engineering.
T. V. Vu, and H. Vahedi, “Distributed energy management for ship power His research interests include nonintrusive load
monitoring synthesis, energy harvesting, and com-
12 munication systems. VOLUME 4, 2016

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2021.3061891, IEEE Access

T. Deeter et al.: Behavioral Modeling for Microgrid Simulation

KATHERINE VASQUEZ received the Naval En-

gineer degree and a Master’s degree in Engineer-
ing and Management from MIT in 2013. She cur-
rently serves as an officer and active duty engineer
in the United States Navy.

BARTHOLOMEW SIEVENPIPER received the

Naval Engineer degree and a Master’s degree in
Mechanical Engineering from MIT in 2013. He
currently serves as an officer and active duty en-
gineer in the United States Navy.

STEVEN B. LEEB received his doctoral degree

from the Massachusetts Institute of Technology in
1993. He has been a member on the MIT faculty
in the Department of Electrical Engineering and
Computer Science since 1993. He also holds a
joint appointment in MIT’s Department of Me-
chanical Engineering. He is concerned with the
development of signal processing algorithms for
energy and real-time control applications.

VOLUME 4, 2016 13

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/