Content uploaded by Sandeep Nagar
Author content
All content in this area was uploaded by Sandeep Nagar on Oct 25, 2014
Content may be subject to copyright.
A Multi-Agent System Incorporating Prediction
Signal to Improve Reliability and Response Time of
a Smart Grid
Sandeep Nagar
School of Engineering
G. D. Goenka University
Gurgaon-Sohna Road,
Gurgaon (Haryana)-India 122103
Email: sandeep.nagar@gdgoenka.ac.in
Ruchi Gupta
School of Electrical, Communication
and Electronics Engineering
Galgotias University
Uttar Pradesh
India
Deependra Kumar Jha
School of Engineering
G. D. Goenka University
Gurgaon-Sohna Road,
Gurgaon (Haryana)-India 122103
Abstract—A Multi Agent System (MAS) architecture based
smart grid is proposed, which includes a prediction signal apart
from status signal to make a decision. Previously proposed MAS
[1] relied only on status signal from various agents and thus
lack the capacity to forecast emergency situation. By including a
prediction signal, which will be based on various environmental
and working parameters of an agent, forecasting of demand
and avialability of power sources can be achieved. The proposed
configuration has been tested using two case studies. The results
confirm that adding a prediction signal will significantly improve
the overall efficiency and reliability of the grid, thus making it
smarter.
I. INTRODUCTION
In contrast to a monolithic architecture, a MAS [1], [2],
[4]–[6] architecture includes many interacting agents to
perform a task. Within the area of grid technologies, agents
can be various types of producers, consumers and distributer.
Similarly a smart grid is an electrical grid which tracks
production, distribution and consumption of electrical energy
and takes intelligent corrective steps based on feedback from
various agents. Smart grid technologies are increasingly
becoming more important for critical facilities where power
failure for even short times can lead to disastrous events.
Smart grid technology are evolving with increasing smartness
in terms of identifying load imbalance and immediately
taking steps to rectify the difference using various agents
available. Research in this arena is driven towards providing
a more efficient, self-healing, reliable [3], and safer and less
constrained solution.
A smart micro-grid can be defined as a low voltage
distribution network with distributed energy resource units,
such as the distributed generation (DG) units and distributed
storage (DS) units and loads. The DG units utilize Diesel
Engines, Micro turbines, Fuel Cells, Photo-voltaic panels,
small wind turbines, and combined heat and power systems.
In a micro smart grid, the capacity of the DG sources varies
from few kW to 1-2 MW. The DS units could be flywheels,
super-capacitors and batteries.
The parameters for judging efficiency of a MAS involves:
1) Accuracy of identifying power deficiency
2) Response time
a) Time taken to determine the appropriate DG/DS,
to fulfil the power requirement
b) Response time to alert DG/DS
c) Time taken to connect DG/DS
3) Reliability of system
Power deficiency identification can be improved by devising
better sensors for the same. Response time can be improved
by devising fast response hardware as well as faster and
efficient algorithm. The algorithms must provide facility for
efficient tracking of power scenario, fast identification and
notification of agent status, fast and reliable connectivity of
power generating and consuming hardware agents. Overall
reliability depends on efficient coordination between agents
and control system. But as long as power status is identified
by real time signals, MAS is subject to failure in case of
emergencies if response time of hardware is less than the time
taken by development of emergency situation. For example,
if weather suddenly changes cloudy and wind speeds drops
drastically then efficiency of photo-voltaic and wind turbines
drop drastically, thus creating a power vacuum suddenly.
Present paper improves upon the architecture provided by
Ruchi et.al. [1] and discuss the role of a prediction signal
(PS) in improving the response and reliability of the smart
grid MAS system. Prediction system makes the smart grid
more self-aware which in turn increases flexibility and hence
reliability. Prediction signal helps in determining projected
power output for a certain duration of time (10 minutes
for present study). With information about projected power
output, decisions about switching to power sources which are
providing ample amount of power or load-shedding can be
taken with greater confidence. To support this new architecture
with concrete evidence, two case studies have been undertaken
and the results support the fact that addition of prediction
signal makes the grid more efficient and reliable.
II. STATUS AND PREDICTION SIGNALS
Status signal (SS) are fed by DG and DS in order to
update the list of power sources which are available for
providing service. Status signals are also fed by consumer
units to describe the power requirements. Based on these
status signals, algorithms are designed which rank the sources
and consumers to provide for power deficiencies.
Performance of DGs like photo-voltaic cells, wind turbines
etc. is function of various environmental factors. For example,
photo-voltaic cells are functional only during well lit day and
their efficiency is a function of age, temperature as well as
weather. Wind Turbine’s performance is also a function of
weather and geographic profile of installation area. Similarly,
DS like battery systems and capacitor banks have a discharge
profile which provides a status signal.
Using only a status signal, makes the MAS unreliable
in emergencies where power vacuum presents itself in
unforgiving manner. It requires frequent updations which
makes the whole system computation power-hungry. It does
not provide an option to generate warning in case of an
upcoming emergency in terms of power generation or power
consumption.
In view of above limitation, an additional prediction
signal (PS) is proposed in the framework of existing control
systems. Prediction based framework can contribute positively
by predicting power requirement, power availability and hence
generate the projected power map for the system with respect
to time and working area.
This makes the system more reliable for situations where
even short power-cuts can be disastrous like emergency units
in hospitals, military installations etc. PS can be further
used to study the performance of the system under certain
environment. This data is useful for artificial intelligence
based systems where systems learn to adapt in case same
situation presents itself in course of operation.
It is important to note that the above-said changes does
not need inclusion of additional hardware. It only requires
addition of date from various prediction models and thus
we are presented with very economical way to achieve
higher efficiency and reliability from existing smart grid
infrastructure.
A. Profile of prediction signals for various agents
Prediction signals from various agents depends on their
functionality. Prediction signal from photo-voltaic agent
(PVAG) includes the variation of efficiency w.r.t shadows of
objects near cell panel, its inclination w.r.t. sun, temperature
of the surroundings, weather like clouds and/or dust storms
can drastically reduce power generation in short span of
time. Dust storm leave the panel with low power generation
conditions for longer time whereas clouds return the clean
day efficiency as soon as they leave the area. Incorporating a
weather prediction data in control system helps in relying on
solar panels for a particular time of the day/year. Prediction
signal for wind-turbine agent (WTAG) also follow weather
profile for a particular region. Prediction system for diesel
generator agent (DGAG) depends on supply of diesel
and maintenance requirements. If DGAG has been used
extensively during first half of the day and diesel supply
is unpredictable then system can lower its reliability on
DGAG and use other agents exhaustively. Prediction signal
for Micro-hydro turbine agent (MHAG) includes water level
forecasting data and operation data for various hydro-turbines
constructed on the route. PS for battery agent (BAG) includes
the data using battery discharge profile which in turn is
function of age, number of charge-discharge cycles, memory
effects and surrounding temperature. PS for Super capacitor
agent (SCAG) includes the discharge profile and temperature
of surroundings.
PS signals can also be charged from consumers based on
their known consumption profile. This requires observing the
consumption pattern over a period of time and modeling the
same. This model gives PS and reliability of model can be
improved by comparing SS and PS. Models can evolve by
learning from comparison of SS and PS by employing machine
learning algorithms.
III. CONTROL SYSTEMS INCORPORATING PREDICTION
SI GNA LS
Existing control systems include status signals (SS) and
generate a rank of suppliers and consumers. Based on the
ranking, suppliers are made available for power requirements.
This existing framework remains unchanged for incorporating
the prediction signal. New ranking table is instead made
using SS and PS both.
Signal profile for SS and PS includes:
TABLE I
SIG NAL PRO FILE
Parameter SS signal profile PS signal profile Comment
Time Local time Local time + x x = 10 min.
Power Immediate Power Projected power output 1
PS involves information about projected power output from
all the agents. This is obtained using equation:
PPO =PP V AG +PW T AG+PD GAG+PM H AG+PB AG+PS CA
(1)
Power from individual agents is calculated using a variety
of factors. PP V AG is complex function of shadow effects
of objects in immediate surroundings during the day time,
operating temperature, weather effects like dust-storms and
cloud formation above the installed area and maintenance
timings. PW T AG depends on wind speed, wind direction and
maintenance timings. PDGAG depends on diesel availability
and maintenance. PM HAG depends upon rainy season profile
and dam routings of water which are installed before the
power station. PBAG depends upon discharge level, charging
time, age, number of discharge cycles, memory effects and
operating temperature. PSC AG differs from PBAG as it
depends on discharge level and operating temperature only
since number of cycles are infinite and charging time is
instantaneous. Real time weather data feed is used to produce
a PS signal based PPO value.
The PPO thus calculated is used to compare with present de-
mand and an emergency situation can be predicted in advance
with an appropriate scheme of dealing with it. Also the system
learn from its own performance gradually as deviation from
ideal behavior leads to error between SS and PS. This error
analysis is used to configure an artificial intelligence based
control system.
IV. ALGORITHM
At a given instant of time SS and PS gives us present power
and projected power after xunits of time (=10 minutes here).
These signals can be used to feed a control algorithm in smart
grid controller (SGC) as shown in Fig. 1.
As illustrated in Fig. 1, the algorithm follows following
steps:
1) Fetch input demand
2) Determine Power available from LDA (Load Distribu-
tion Agent) from SS and Projected power output from
PS
3) Generate producer list
4) Add producer in selected producer list till the demand
met or all producer got added in the list.
5) Send the list to SGC (Smart Grid Controller)
6) Send load shedding message to SGC if complete demand
not met.
PPO determines the ranking of a generator unit. When
environmental factors force a producer to drop its output, its
sensed 10 minutes in advance and another producer immedi-
ately gets a higher rank so that power source can be switched
to a more reliable producer in those particular circumstances.
Fig.2 shows an algorithm for this purpose and it follows the
following steps:
1) Fetch a producer from the list
2) Generate new list for SGC of selected producer in case
PPO of any producer changed.
3) Else check PPO of next producer
4) This routine will run in closed loop
Deviation of values of power output from PPO can be used
to test the reliability of a system. This study is beyond the
scope of present paper and it is proposed that a simulation
study based on our algorithm can be undertaken for various
configurations of smart-grid, to evaluate their reliability and
efficiency.
Fig. 1. Flow chart depicting algorithm used for SGC
Fig. 2. A closed loop algorithm for ranking producers by comparing present
power output (P) and projected power output (PPO)
V. SIMULATION AND RESU LTS
Power generation sources have specific startup times before
giving stabilized power. This time varies depending upon the
sources and can be up to few minutes. In view of this, an
additional prediction signal (PS) can always be handy as it
gives system operator an extended time to plan and execute
the system operation in a more effective way.Prediction based
control system enables stronger communication between var-
ious agents and thus helps improve the reliability matrix.
Fig. 3. Demand power of LAG (Load Agent) for 24 hrs
The prediction process and its validity are explained through
the following case scenarios. Demand of the LAG as shown
in Fig. 3 is provided by Ruchi et.al [1]
A. Case 1: Day time and power is not available from WTAG
Fig. 4. Showing unavailability of power by WTAG
Fig.4 shows the load demand during the day time. As the
demand increases and since per unit price from WTAG is
less as compared to from Grid, SGC can make a decision
to shift the power supply using WTAG. But now assuming
that weather changes to adverse conditions to performance
of WTAG, like less or unpredictable wind speeds and vary-
ing wind directions. WTAG informs SGC about the cloudy
weather (predicted well in advance, here 10 minutes). As the
grid is still available along with other agents, SGC updates the
ranking of supply agents and puts WTAG to lowest priority.
When WTAG goes down now, power vacuum of 300 KW
would have been created suddenly if hadn’t been predicted
and informed to SGC. Hence, demand management becomes
flexible.
B. Case 2: Day time and power is not available from PVAG
Fig. 5 shows the load demand during day time. When
the grid goes off due to power shedding SGC fetch power
Fig. 5. Showing unavailability of power by PVAG
from WTAG, MHAG, BAG, PVAG and DGAG based on the
per unit price. Now suppose PVAG is not available at this
moment due to dust storm, PVAG sends a signal to SGC about
low generation conditions (predict well in advance, here 10
minutes). If PVAG had not predicted about its low generation
conditions a power vacuum of 350 KW would have been
created suddenly affecting critical loads like hospitals etc. In
present case SGC takes an immediate decision to fulfill the
demand from DGAG. If the supply is more than demand, SGC
takes the decision of not taking power from any other DGAG.
In this way system reliability is improved which reduces the
economic losses incurred by consumers when power is lost.
To improve the power system reliability i.e. to provide
backup power, in case of power interruption and to prevent
the spikes in case of transition SCAG uses the capacitor bank
as energy storage device and tries to maintain the frequency.
In the above two cases it is assumed that load generation is
fixed i.e. there is no variation in the load demand. When there
is a continuous rise in demand, LAG alerts SGC for a pos-
sible overload condition. High speed under-frequency relays
installed at strategic locations sense the overload condition and
initiate load shedding in order to keep system frequency to an
acceptable level. If the grid is available then excess demand
can be fulfilled but if grid is not available then SGC takes the
decision to fulfill demand of critical loads and low priority
load can be disconnected temporarily from the system. In this
way, the introduction of prediction signal along with status
signal makes the grid smarter.
VI. CONCLUSION
An algorithm for controlling a smart grid based on MAS
architecture, has been proposed. Here both SS and PS has
a stake to efficiently control the emergency situations by
switching to power producers which are reliable in near
future. Prediction signal is proposed to be composed of sum
total of increase/decrease of power output as a function of
various environmental factors. Two case studies have been
undertaken to test the algorithms and it has been found that
addition of prediction signal enables the system to update the
priority list of power suppliers well in advance and prepare for
emergencies. Present scheme also paves the path to learn the
deficiencies in simulation models for environment and evolve
them.
REFERENCES
[1] R. Gupta, D. Kumar Jha, V. Kumar Yadav and S. Kumar, ”A Multi-Agent
Framework for Operation of a Smart Grid,” Energy and Power Engineer-
ing, Vol. 5 No. 4B, 2013, pp. 1330-1336. doi: 10.4236/epe.2013.54B252.
[2] Dimeas, AL.; Hatziargyriou, N.D., ”Operation of a Multiagent System
for Microgrid Control,” Power Systems, IEEE Transactions on , vol.20,
no.3, pp.1447,1455, Aug. 2005 doi: 10.1109/TPWRS.2005.852060
[3] Moslehi, K.; Kumar, R., ”A Reliability Perspective of the Smart Grid,”
Smart Grid, IEEE Transactions on , vol.1, no.1, pp.57,64, June 2010 doi:
10.1109/TSG.2010.2046346
[4] Fangxing Li; Wei Qiao; Hongbin Sun; Hui Wan; Jianhui Wang; Yan Xia;
Zhao Xu; Pei Zhang, ”Smart Transmission Grid: Vision and Framework,”
Smart Grid, IEEE Transactions on , vol.1, no.2, pp.168,177, Sept. 2010
doi: 10.1109/TSG.2010.2053726
[5] Dejia Shi; Li Wang; Jing He, ”Notice of Violation of IEEE Publication
Principles The Design of Multi-agent System in IDAPS Microgrid,”
Intelligent Interaction and Affective Computing, 2009. ASIA ’09. Inter-
national Asia Symposium on , vol., no., pp.63,66, 8-9 Dec. 2009 doi:
10.1109/ASIA.2009.17
[6] Colson, C.M.; Nehrir, M.H., ”Comprehensive Real-Time Microgrid
Power Management and Control With Distributed Agents,” Smart Grid,
IEEE Transactions on , vol.4, no.1, pp.617,627, March 2013 doi:
10.1109/TSG.2012.2236368
