Energy and Power Engineering, 2013, 5, 1053-1059
doi:10.4236/epe.2013.54B201 Published Online July 2013 (http://www.scirp.org/journal/epe)
Multi-Layer Optimization for Load Scheduling to Manage
Unreliable Grid Outages in Developing Countries
Amit S. Closepet
Spectrum Consultants, Bangalore, India
Received 2013
ABSTRACT
This paper describes the significant cost saving opportunities for consumers in developing countries by the use of com-
putational intelligence and demand-side-management techniques to mitigate the massive use of diesel back-up during
grid outages. Application of load scheduling optimization is investigated during scheduled power outages, for residen-
tial consumer in India. The specific load shifting approaches explored include a day ahead predicted load schedule
which is generated by performing a DSM referring to the forecasted day ahead outage. Whereas in reality the predicted
may not match the actual outage, thus in these cases a fuzzy logic rule base is referred on real time basis to take correc-
tive action & reach the best optimal load schedule possible to attain the lowest cost. The load types modeled include
passive loads and schedulable, i.e. typically heavy loads. It is found that this multi-level DSM schemes show excellent
benefits to the consumer. The maximum diesel savings for the consumer due to load shifting can be approximately
ranging from 45% to as high as 75% for a flat-tariff grid. The study also showed that the actual savings potential de-
pends on the timing of power outage, duration and the specific load characteristics.
Keywords: Residential Demand Side Management; Diesel Mitigation; Real Time Load Shifting; Power Outage
Management; Computational Intelligence
1. Introduction
Demand-side-management (DSM) policies are being
formulated by various stakeholders in India and other
developing countries. These policies are specifically tar-
geted to overcome large energy demand-supply gaps, to
provide inclusive and reliable power for entire popula-
tions. For example, in India, load scheduling has recently
been implemented successfully for the agricultural sector.
As in developed countries, load scheduling is driven by
the utility for peak clipping of demand, load shifting for
energy conservation and/or supporting load growth. In
this work, our aim is to highlight the urgent need for de-
mand-side-management policies to address one of the
major unaddressed challenges for a consumer in a de-
veloping country which is the problem of frequent power
outages. DSM solutions and policies need to be devel-
oped, validated and framed to enable the consumer get
reliable power and reduce his dependence on expensive
diesel back-up systems. The highly stochastic nature of
the power grid (Figure 1) in developing economies with
relentless power cuts forces consumers to rely heavily on
diesel back-up systems for business continuity, risk mi-
tigation and load execution. Major cost investments are
required for both installation and operation diesel
back-up systems. Recent articles [1-4,17] have shown
that, in some cases, cost can run as much as high as 50%
of the customer’s annual power budget. In this study, we
explore the use of demand management techniques that
will enable the consumer mitigate power outages, cut the
dependence on diesel back-up and provide significant
cost savings to the consumer.
Tables 1(a)-(b) in [18] provide power outage data for
several major cities in India and in other developing
economies such as Africa, Philippines, South America
and observe that power outages range from 2 hours to
more than 10 hours a day. Power blackouts typically re-
sult in total loss of power to large parts of the entire city
to large districts. Different scenarios and causes for
planned/unplanned power outages in India are described
in [4,5]. Due to this grid unreliability, the market for
genset and UPS systems is India is worth several billion
dollars and growing at a rapid 20% annual rate. In India,
Figure 1. Stochastic grid in developing countries.

A. S. CLOSEPET
1054
residential consumers sometimes pay large premium
(~3X) over grid power due to use of expensive back-up
systems [4,5]. Service businesses (e.g. photocopier cen-
ters, medical diagnostic labs, service apartments, wed-
ding halls etc.) charge higher rates to the consumer dur-
ing power outages. Home owners association (HOA) in
apartment complexes are faced with large diesel bills due
to a shared gen-set and these additional costs are peri-
odically collected from the consumer. In hospitals, the
continuous power is ensured by using UPS systems. Ac-
cording to the Bureau of Energy Efficiency in India [6],
to deliver a sustained economic growth rate of 8% to 9%
through 2031-32 and to meet life time energy needs of all
citizens, India needs to increase its primary energy sup-
ply by 3 to 4 times and electricity generation capacity
about 6 times. Based on these aforementioned statistics
and the present high inefficiencies in the grid, it is likely
that power outages are here to stay unless DSM policies
are directly targeted at mitigating the large power out-
ages for all consumers.
We explore two specific DSM techniques manage
outages for a residential load in India. Our goals are to
minimize the quantum of loads executed during outage
by load-shifting and deliver costs savings by significantly
reducing the consumption of diesel. Decisions to whether
to execute the load with diesel or to compromise load the
altogether are often encountered for the emerging market
consumer. Specific power outages are announced well in
advance by the utility publicly for each specific locality.
For example, in the city of Chennai in South India, the
local utility has recently announced a 2 hour power out-
age daily for the months of March to June 2012.
In this study, application of load scheduling optimiza-
tion is investigated during scheduled power outages, for
residential consumer in India. The specific load shifting
approaches explored include a day ahead predicted load
schedule which is generated by performing a DSM refer-
ring to the forecasted day ahead outage. Whereas in real-
ity the predicted may not match the actual outage, thus in
these cases a fuzzy logic rule base is referred on real time
basis to take corrective action & reach the best optimal
load schedule possible to attain the lowest cost. The load
types modeled include passive loads and schedulable, i.e.
typically heavy loads. It is found that this multi-level
DSM schemes show excellent benefits to the consumer.
Implementation of a load scheduler can be extremely
difficult for a consumer in a developing country. The
reasons are many including ease of use, availability of
controllable loads etc. In India, consumers typically
switch off their heavy loads during a power outage and
execute them after power is restored. In newer apartment
buildings, the heavy load lines are usually a separate cir-
cuit (e.g. 15 amps) and the apartment back-up generator
simply doesn’t provide power to these lines. All heavy
and shiftable loads are often connected to the 15 amp line.
This technique of casually rescheduling heavy loads re-
sults in unexpected peaks for the utility as soon as power
is restored. While the actual implementation of the load
control and scheduling can be accomplished either by the
utility or the end consumer themselves, the aim of the
DSM policy needs to be consumer-centric. In other
words, the consumer needs to always have the flexibility
of load selection and execution without yield controlling
to the utility.
1.1. Literature Survey
Numerous studies have focused on the impact of unreli-
able grids on consumer cost as seen in the following
works: An in-depth investigation into the impact of pow-
er outages for consumers and businesses in Africa is
performed in [4]. This study also assesses the economic
consequences of the unreliable grids. A report on real
power cost in India [5] reveals that the overall intent of
providing cheap and affordable power to the consumers
in the country is noble, but if the supplies are inadequate
or unreliable, the consumers could actually end up pay-
ing a much higher price. A report from United Nations [6]
provides directions to expand access of modern energy
services at the household level. An application of com-
bined model of extrapolation and correlation techniques
for short term load forecasting of an Indian substation is
presented in [7]. Specific opportunities for DSM in the
Indian scenario are presented in [8]. Low-cost energy
generation using bio- mechanical energy is presented in
[9] and this provides a technology options for both off-
grid users as well as on-grid users who have unreliable
power. The use of casual scheduling loads with time-
varying prices using stochastic dynamic programming is
studied in [11] and its effect on consumer cost are re-
ported.
In [13], a power scheduling protocol for demand re-
sponse in smart grid system is explored which focuses on
limiting the allowable power loads. Algorithmic en-
hancements to a scheduler for residential DSM are pre-
sented in [15].
1.2. Simulation Approach and Analysis
Methodology
The MATLAB methodology to model the demand side
management optimization and scheduling are described
in this section. The MATLAB code is structured in such
a manner that it fetches all the input data from various
excel files, these excel files can be edited for demand, for
the individual load power characteristics, for the load
start time, the load run time, for forecasted outage start
time & outage duration etc., Once these inputs are ready
we can go to the MATLAB GUI to run the code.

A. S. CLOSEPET 1055
Once the code is run for a forecasted outage it results
in a new load schedule for the following day depending
on the outage. Due to the unreliable grid, we have as-
sumed an error in the outage scenario of maximum of 1
hour on either sides of the forecasted outage. Thus to
simulate this unreliable grid we do a real time fuzzy logic
based DSM on the loads by creating an error in the out-
age either in the outage start time or outage end time or
even both. Thus depending on the actual outage the fuzzy
logic rule base is referred for a further correction in the
load schedule to reach to the best optimal cost.
The baseline costs assumed for the grid is 5c/KWhr
(residential) and the baseline diesel costs assumed in the
simulation are 20 c/KWhr. As the power characteristics
of the loads are not constant we have divided the day into
multiple of a 5 minutes chunk, so 24 hours is considered
as 288 chunks, by doing this we can be very accurate in
calculating the effective cost, for a better and simple un-
derstanding we have assumed all the heavy loads consid-
ered in the paper i.e. 3 Geysers, 1 washing machine, 1
dishwasher & a dryer to have flat power characteristics
curves.
1.3. Optimization Equations
The cost minimization equation is as follows: Input pa-
rameters
LPASSIVE(t) Passive Load at time t
LSHIFT,i(t) Shiftable Load i at time t
Total Load ∑(L(t)) = ∑LPASSIVE(t) + ∑LSHIFT ,i(t) for
i=1,n tGGrid available time for a day
tBDiesel usage time in a day CG(tG) Cost per unit
with grid CB(tB) Diesel cost per unit
Total cost per unit at time t C(t)
= CG(tG) + CB(tB)
Total Cost CTOTAL = ∑C(t)L(t) for t = 0,24
COPT = Min (CTOTAL)
Algorithm: - Finding optimal load schedule
The above is done by following the below mentioned
flow chart in Figure 2.
Figure 2 Basic algorithm flow chart
L1NrsT Normal Start Time Of Load (1)
L1NST Earliest Start Time Of Load (1)
L1LST Latest Start Time Of Load (1)
L1RT Run Time Of Load (1)
L1NrETNormal End Time Of Load (1)
L1NrET= L1NrsT+ L1RT
Figure 3. Load schedule constraint.
Figure 3 speaks about how every shiftable load is
normally scheduled & also what are its constraints i.e.
the load cannot be shifted randomly during the day but
has an earliest start limit & also a latest start limit.
Hence any shifting of these loads has to be done be-
tween this time frame.
Figure 4. Expected outage for the following day.
bExST Expected Outage Start Time
bExSTExpected Outage End Time
In Figure 4 we can clearly observe that the outage is
expected to affect the load, thus this load has to be
shifted, Thus the load is scheduled to a new start time
either before the outage or after the outage, this com-
pletely depends on the load constraints & also the run-
time of the load, if the gap is available on both sides of
outage, the algorithm chooses to shift the load before the
outage as it is safer to execute the load beforehand, rather
to risk the execution of the load with the unreliable grid
supply.
This is seen in Figure 5.
Figure 5 Day ahead Load Shift Sc he dule .

A. S. CLOSEPET
1056
L1NeSTNew Expected Load Start Time
L1NeET New Expected Load End Time
L1NeET= L1NeST+ L1RT
Figure 6. Actual Outage.
bAcST Actual Outage Start Time
bAcETActual Outage End Time
In Figure 6 we can observe that the actual outage is
overlapping the new scheduled start time of the load, this
data is obtained from the real time supply sensors.
Now the real time Fuzzy Logic Rule base for this kind
of a scenario where a second shift of the load is required
comes into action.
The rule mentioned below comes into action & the
load is shifted at the bAcET., as shown in Figure 7.
If bAcST ≤L1NeST & L1NeST ≤ bAcET≤ L1NeET =>2nd Shift
Figure 7. Actual Load Schedule.
L1AcST  Actual Load Start Time
Thus the L1AcST= bAcET
All the shiftable loads undergo this exercise to get the
best possible optimal solution to attain the least cost, by
finally reaching COPT = Min (CTOTAL)
2. Results
In this section, key results and benefits from the MAT-
LAB Tool for the 2 level optimization of load-scheduling
of residential loads for diesel mitigationare discussed.
Through the forecast the expected outage can be gen-
erated, thus this expected outage will lead to a first level
demand side management & hence an expected new load
schedule.
Table 1 shows the forecasted diesel savings of various
different outage durations starting at 2 different peak
times in a day. For the case of a 2 hours outage starting at
1pm the expected diesel savings of as high 65.95% can
be seen.
The following Figure 8 explains the aggregate refer-
ence load curve i.e. the original load curve, here we can
see the 2 peaks prominently. The break-up of the load
will be seen in Figure 9.
Next the Figure 10 explains the expected outage lead-
ing to a load shift and finally a new expected load sched-
ule for the next day .
Comparing Figure 11 & Figure 9 we can clearly no-
tice that for an expected outage of 1pm to 3pm the first
level DSM will shift the dishwasher & dryer to a new
time & will schedule them to start at 3pm.
Table 1. Forecasted Diesel Percentages For Various Out-
ages.
Forecasted Diesel Percentage Savings
Start time/ Outage duration 2 hours 4 hours 6 hours 8 hours
7PM 59.80 52.79 54.41 58.60
1PM 65.95 53.52 35.58 29.62
Figure 8. Origin al Day Ah eadLoad C urve.
Figure 9. Individual load schedule.
Figure 10. Expe cted load schedu le.

A. S. CLOSEPET
1057
shift, this can be observed at an aggregate level in Figure
12 & at an individual level in Figure 13.
Due to the unreliable grid there could be many sce-
narios that are very close to the forecasted outage. Table
2 shows 25 different cases which are very close to the
expected outage.
Here we have assumed a maximum of 1 hour error in
the outage on either sides of the expected outage.
The last column in Table 1 shows the type of shift, i.e.
the green colour shows that those outages are leading to a
real time second shift. The pink colour shows that these
outages are not interfering the first load schedule &
hence need no real time shifting.The Figure 12 shows
the variation in the outage from the expected outage, In
this particular case the actual outage starts at 12pm to
3:30pm.This case involves a real time fuzzy logic based Figure 11. Expected individual load
schedule.
Table 2. Shows 25 different actual cases around a
single expected 2 hours outage case.
Table for 2 Hours Blackout
SL NO TYPE OF CASE TIME OF BLACKOUT % SAVINGS TYPE Of SHI
1 ACTUAL CASE 12PM TO 2PM
75.0
2 ACTUAL CASE 12PM TO 2:30PM
70.9
3 ACTUAL CASE 12PM TO 3PM
67.8
4 ACTUAL CASE 12PM TO 3:30PM
65.4
5 ACTUAL CASE 12PM TO 4PM 63.5
6 ACTUAL CASE 12:30PM TO 2PM
55.3
7 ACTUAL CASE 12:30PM TO 2:30PM
53.8
8 ACTUAL CASE 12:30PM TO 3PM
52.8
9 ACTUAL CASE 12:30PM TO 3:30PM
52.1
10 ACTUAL CASE 12:30PM TO 4PM
51.5
11 ACTUAL CASE 1PM TO 2PM
75.0
12 ACTUAL CASE 1PM TO 2:30PM
69.6
13 EXPECTED CASE 1PM TO 3PM 66.0
14 ACTUAL CASE 1PM TO 3:30PM
63.3
15 ACTUAL CASE 1PM TO 4PM
61.3
16 ACTUAL CASE 1:30PM TO 2PM
75.0
17 ACTUAL CASE 1:30PM TO 2:30PM
66.0
18 ACTUAL CASE 1:30PM TO 3PM 61.3
19 ACTUAL CASE 1:30PM TO 3:30PM
58.4
20 ACTUAL CASE 1:30PM TO 4PM 56.5
21 ACTUAL CASE 2PM TO 2:30PM
46.7
22 ACTUAL CASE 2PM TO 3PM 46.7
23 ACTUAL CASE 2PM TO 3:30PM 46.7
24 ACTUAL CASE 2PM TO 4PM
46.7
Figure 13. Fuzzy lo gic base d real t ime optimi zed individ ual
load
curve.
Figure 12. Fuzzy logic based r eal tim e optimiz ed load curve.

A. S. CLOSEPET
1058
Table 3. Shows 25 different actual cases ar ound a single expected 4 hours outage case.
Table for 4 Hours Blackout
SL NO TYPE OF CASE TIME OF BLACKOUT % SAVINGS TYPE OF SHIF
1 ACTUAL CASE 12PM TO 4PM 63.51
2 ACTUAL CASE 12PM TO 4:30PM 59.79
3 ACTUAL CASE 12PM TO 5PM 56.63
4 ACTUAL CASE 12PM TO 5:30PM 54.34
5 ACTUAL CASE 12PM TO 6PM 52.22
6 ACTUAL CASE 12:30PM TO 4PM 51.71
7 ACTUAL CASE 12:30PM TO 4: 30PM 48.28
8 ACTUAL CASE 12:30PM TO 5PM 45.02
9 ACTUAL CASE 12:30PM TO 5: 30PM 43.48
10 ACTUAL CASE 12:30PM TO 6PM 41.67
11 ACTUAL CASE 1PM TO 4PM 61.29
12 ACTUAL CASE 1PM TO 4:30PM 57.14
13 EXPECTED CASE 1PM TO 5PM 53.52
14 ACTUAL CASE 1PM TO 5:30PM 51.01
15 ACTUAL CASE 1PM TO 6PM 48.72
16 ACTUAL CASE 1:30PM TO 4PM 56.52
17 ACTUAL CASE 1:30PM TO 4:30PM 51.49
18 ACTUAL CASE 1:30PM TO 5PM 49.10
19 ACTUAL CASE 1:30PM TO 5:30PM 44.44
20 ACTUAL CASE 1:30PM TO 6PM 41.94
21 ACTUAL CASE 2PM TO 4PM 46.67
22 ACTUAL CASE 2PM TO 4:30PM 40.58
23 ACTUAL CASE 2PM TO 5PM 35.90
24 ACTUAL CASE 2PM TO 5:30PM 32.94
25 ACTUAL CASE 2PM TO 6PM 30.43
From Table 2 we can see that this real time shifting of
loads will lead to a diesel savings percentage of 65.4%.
Here we can observe that the real time shift of washing
machine, dryer & dishwasher which was previously
scheduled to start at 12pm, 3pm & 3pm respectively will
now start at 3:30pm.
Similarly Table 3 shows the actual 25 cases that are
possible around an expected outage of 4 hours starting at
1pm. The expected diesel savings from Table 1 is
53.52% where as in reality due to the unreliable grid the
savings percentage can vary from 30.42 % to 63.51%.
3. Conclusions
Use of multi-layer load-shifting techniques to mitigate
power outages in developing countries shows significant
cost savings potential by massive reduction in diesel
consumption by load-scheduling. The maximum diesel
reduction for the consumer due to load shifting during
power outages can be approximately 45% to 75% for a
flat- tariff grid. The study also showed that the actual
savings potential depends on the timing of power outage,
duration and the specific load characteristics. As diesel
prices increase, the economic benefits of load-shifting
are also increase correspondingly. For blackouts of lesser
duration (e.g. 2 hrs) the benefits in saving diesel can be
as much as 75%. For longer blackouts (e.g. 8 hours), the
diesel savings is in the range of 20%-60%. DSM policies
for developing countries should consider specific ap-
proaches to mitigate power outages and provide relief to
customers. Clearly, challenges exist in implementation of
DSM policies since most consumers in India and frugal
markets have outdated appliances that are unintelligent
with a severe need to develop low-cost smart net-
work-controllable solutions as a retrofit.
REFERENCES
[1] Power Annual report 2009-10, CES, Govt. India.
[2] “Energy and the Millennium Development Goals in Af-
rica”, 2006, UN World Summit.
[3] Development Goals in Africa Electricity Supply Moni-

A. S. CLOSEPET 1059
toring Initiative (ESMI), Prayas, India.
[4] V. Foster and J. Steinbuks, “Paying the Price for Unreli-
able Power Supplies” In-House Generation of Electricity
by Firms Africa, 2009
[5] “Real cost of power” 2009, Universal Consulting.
[6] “Energy poverty” World Energy Outlook for UN General
Assembly on Millennium Goals 2010
[7] R. Behera, B. Panigrahi and B. Bhusan, “Hybrid Short
Term Load Forecasting Model of Indian Grid” Energy
Power Engg, 2011
[8] S. Padmanaban and Ashok “Electricity DSM in India
Strategic & Policy Perspective”.
[9] V. Natarajan, A. Baxi, R. Padmanabhan and V. Magesh-
kumar, “Low-Cost Bio Mechanical Energy Generator for
Off-Grid Users in Emerging Markets,” 2011 IEEE
GHTC.
[10] Report, “Action Plan for Energy Efficiency,” 2012, Bu-
reau Energy Efficiency, India.
[11] T. Kim and V. Poor, “Scheduling Power Consumption
with Price Uncertainty,” 2011, IEEE Trans Smart Grid.
[12] Amir-Hamed Mosenian-Rad and A. Leon-Garcia, “Opti-
mal Residential Load Control With Price Prediction in
Real-Time Electricity Pricing Environments,” 2010, IEEE
Trans. Smart Grid.
[13] M. A. A. Pedrasa, T. D. Spooner and I. F. MacGill,
“Scheduling of Demand Side resources Using Binary
Particle Swarm Optimization,” IEEE Transactions Power
System, Vol. 24, No. 3, 2009.
doi:10.1109/TPWRS.2009.2021219
[14] K. Herter, “Residential Implementation of Critical-peak
Pricing of Electricity,” Energy Policy, Vol. 35, 2007.
doi:10.1016/j.enpol.2006.06.019
[15] M. A. A. Pedrasa, Ted D. Spooner and Iain F. MacGill,
“Coordinated Scheduling of Residential Distributed En-
ergy Resources to Optimize Smart Home,” IEEE Trans-
actiond Smart Grid, Vol. 1, 2010.
[16] “Energy Infrastructure in India,” 2011, World Bank
Country Report.
[17] V. Natarajan and A. S. Closepet, “Statistical Analysis of
Cost of Energy Due to Power Outages in Developing
Countries,” 2012, Future Computing Conf Submission,
France.
[18] V. Natarajan and A. S. Closepet, “Demand-Side Ap-
proaches To Manage Electricity Outages In Developing
Countries,” 2012, Energycon, Florence, Italy.