InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013

DOI:10.14355/ijes.2013.0305.03

www.ijesci.org

OptimumDecisionPolicyForReplacementof
ConventionalEnergySourcesbyRenewable
Ones
MaryamParsa1,NasirUddinAhmed*2,MustaphaC.E.Yagoub3
SchoolofElectricalEngineeringandComputerScience,UniversityofOttawa,ON,Canada
mparsa@uottawa.ca;*2ahmed@site.uottawa.ca;3myagoub@site.uottawa.ca

Abstract

DecisionPolicy;ConventionalandRenewableEnergySources

Withtheincreaseofworldpopulationandindustrialgrowth
of developing countries, demand for energy, in particular
electric power, has gone up at an unprecedented rate over
the last four decades. To meet the demand, electric power
generation by use of fossil fuel has increased enormously
thereby producing increased quantity of greenhouse gases
contributingmoreandmoretoatmosphericpollutionwhich,
climate Scientists believe, can adversely affect the global
climate,andhealthandwelfareofworldpopulation.Inview
of these, there is global awareness of looking for alternate
sources of energy such as natural gas, hydropower, wind,
solar, geothermal and finally biomass etc. It is recognized
thatthisrequiresreplacementofexistinginfrastructurewith
newsystems,whichcannotbeachievedovernight.Optimal
controltheoryhasbeenwidelyusedforthelastfivedecades
indiverseareasofphysicalsciences,medicine,engineering,
economicsandsocialsciences.

Introduction
With the increase of world population and
unprecedented industrial growth in many populous
countries,demandforenergy,inparticularelectricity,
has been steadily increasing over last 3040 years. To
meetthiseverincreasingdemand,useoffossilfuelfor
generationofelectricityhasbeenrisingandalongwith
that the production of toxic gasses potentially
threateningtheclimateandpublichealth.Ontheother
hand, the conventional energy sources such as Coal
and Petroleum are becoming more and more scarce
althoughtherearescientistswhobelievethatthereare
still ample supply of these sources (Shafiee, Topal,
2009).Mostoftheresearchersthinkthatthesesupplies
are transient and will be depleted within next sixty
years(Steinberger,2009).Inanycase,consideringjust
the adverse effects of pollution on health and global
climate, it has become mandatory for industrial
countries to seek for alternate sources and follow a
policy of gradual replacement of polluting power
sources by clean and renewable ones without
adverselyaffectingtheeconomy.

The main motivation of this article is to use this theory to

find the optimum strategy (decision or policy) for
integration of all currently available renewable energy
sources with the existing electric power generating systems
with the ultimate goal of elimination of fossil fuel. Eight
main energy sources such as Coal, Petroleum, Natural Gas,
ConventionalHydro,Wind,Solar,GeothermalandBiomass
areconsideredinadynamicmodel.Thestateofthedynamic
modelrepresentsthelevelofpowergenerationfromeachof
thesourcesatanytimet.Theproposedobjectivefunctionis
based on the desired target level of power generation from
each of the available sources at the end of the plan period
while reducing the production of greenhouse gasses.
Pontryagin Minimum principle is used to determine the
optimal control or decision policy. Official released data
fromtheU.S.EnergyInformationAdministrationisusedas
a case study. Based on this data and a mathematical model
proposed in a paper (Miah, Ahmed and Chowdhury, 2012)
published by Energy Economics combined with the
minimum principle, an optimal policy is presented for
integration of renewable energy sources to the national
powergrid.

In this paper, we consider all eight main energy

sources used for generation of electricity. These are
Coal, Petroleum, Natural Gas, Hydro, Wind, Solar,
Wood and Biomass and Geothermal. Our choice of
these sources is based on the availability of
comprehensive official released data of U.S. Energy
Information Administration (U.S. EIA). This data is
used along with optimal control (decision) theory
(Ahmed,1988) to develop a methodology for optimal
reduction or replacement of fossil fuel by renewable
energysources.
Inarecentpaper(Miah,AhmedandChowdhury,2012)
LotkaVolterra model was proposed for use as a
dynamic model of power generation involving only

Keywords
Optimization; Mathematical Models; Optimal Control; Optimum

311

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

level of electricity generated from Solar Power,

(includingthermalandphotovoltaic),x7(t)isthelevel
generated from Geothermal Source and finally x8(t) is
the level of generation from Wood and Biomass. The
vector

two sources such as Renewable and Conventional

sources.Weuseasimilarmodelbutforeightdifferent
sourcesandusetheavailableofficialdata(U.S.EIA)to
validate the model and then develop optimal
generation policy based on the Pontryagin minimum
principle(Ahmed,1988).TheLotkaVolterrasystemof
nonlinear differential equations (Ahmed, 1988, p24)
can be used for modeling cooperative as well as
competingagentsinanecologicalenvironmentaswell
asinmarketeconomy.

x(t ) ( x1 (t ),..., x8 (t ))' , t 0

denotesthestate(level)ofpowergenerationfromeach
ofthesourcesasafunctionoftime.
In the equation (1), ui, i = 1 8 represents the
percentageincreaseordecreaseofgenerationfromthe
ith source. The parameter ij describes the impact of
generationlevelofthejthpowersourceonthegrowth
(or decline) rate of generation level of the ith power
sourceandsimilarlyjiistheimpactofthegeneration
level of the ith power source on the growth (decline)
rate of generation level of the jth energy source. Note
that in general ij ji. In other words, the matrix of
interaction is not necessarily symmetric. There are
fiftysixinteractionparametersinthemodelproposed
for our system and they must be identified before
undertaking policy optimization. This will be
consideredintheSystemIdentificationsection.

Restofthepaperisorganizedasfollows.Thedynamic
modelofthesystemandtheproposedcostfunctional
are presented in section II. This is followed by the
system identification problem (also called inverse
problem) and the problem concerning optimum
decision policy in sections III and IV respectively.
Different optimization problems are discussed in
section V. Numerical algorithms and data are
presented in sections VI and VII respectively. The
numerical results are presented in section VIII. The
paperisconcludedinsectionIX.
Problem Description and Dynamic Model of
the System

PerformanceorCostFunction
The primary goal of this paper is to develop a
methodology for gradual and optimal replacement of
the polluting energy sources, such as Coal and
Petroleum, by clean and renewable energy sources,
such as Natural Gas, Wind, Solar, Geothermal and
Biomass. The objective function to be chosen must
incorporate the main concerns of the planner. For
example, one may be interested to meet the target
levelofgenerationattheendoftheplanperiodwhile
reducing the cost of implementation and greenhouse
gasemissionduringtheplanperiod.

LotkaVolterraModel
The dynamics of biological ecosystems are basically
described by a system of LotkaVolterra equations
(Ahmed,1988,p24),(Goel,MaitraandMontroll,1971).
Basically this model can be used in both cooperative
and competitive environment. More recently, this
model has been used also in the study of energy
economics (Miah, Ahmed and Chowdhury, 2012). In
the study of energy economics, there are many
competing sources for power generation and hence
this model can be suitably modified to model the
energygenerationmarketenvironment.

The following general expression given by equation

(2), called the cost function, can be used to include
manysuchobjectives.

Fortheproblemconsideredinthisarticle,asystemof
eight firstorder nonlinear differential equations is
introducedasfollows:
xi ui xi

j 1, j i

i j xi x j

tf

J (u )

(2)

Thecostfunctiongivenbyequation(2)consistsoftwo
parts representing the running cost and the terminal
cost. The running cost may include the
implementation cost and the penalty for producing
greenhouse gasses during the plan. The terminal cost
represents the mismatch between the desired level of
generation and the actual level of generation reached
bythesystemattheendoftheplanperiod.

(1)

where xi, i = 1 8 denotes the level of power

generated by the ith source at any given time. In
particular,x1(t)isthelevelofelectricpowergenerated
from Coal at time t, x2(t) is the level of electricity
generatedfromPetroleum,x3(t)isthelevelofelectricity
generated from Natural Gas, x4(t) is the level of
electricity generated from Conventional Hydro, x5(t) is
thelevelofelectricitygeneratedfromWind,x6(t)isthe

312

(t , x(t ), u (t ))dt ( x(t f ))

The objective is to determine the decision (or control)

policy u as a function of time, taking values from the

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

www.ijesci.org

set U={u: |ui| 1, i = 1 8} so that it minimizes the

costfunctionJ(u).Todeterminetheoptimumdecision
policy, we need the relevant mathematical theory of
optimalcontrolandsystemidentificationpresentedin
thefollowingsections.

The optimum solution of the problem considered in

thisarticle,iscontainedinthefamilyoffunctions{uo,
xo, } satisfying all the above equations called the
extremals.Laterintheumericallgorithmsection,
thewaytheseequationsareusedisdiscussed.

Pontryagin Minimum Principle and Optimal

Control Theory

System Identification Problem

As stated before, there are fifty six unknown
parameters in the model chosen for this problem
Equation (1) that provides the interaction (coupling)
between generationlevels of different types of power
sources. It is interesting to note that optimal control
theory can be used also for identification of
parameters(ij,1i,j8).Giventhaty(t),tI,is
thenaturalresponseofthesystem,andx(t,),tI,the
responseoftheproposedsystemmodel(1),ameasure
of mismatch can be introduced by the following
functional

Pontryagin Minimum Principle is a powerful tool for

dynamic optimization and so it is clearly suitable for
the proposed problem of this article. For detailed
theory see (Ahmed 1988, Kirk 2004). According to
(Ahmed 1988, Kirk 2004, Moss and Kwoka 2010,
Benson and Franklin 2008) the Hamiltonian function
playsanimportantroleinthismethodofoptimization.
Consider a general ndimensional system with state
equation(n=8inourproblem)
x f (t , x, u ), x(0) x0

(3)

J ( )

andtheobjectivefunctional(2)initsgeneralform.The
Hamiltonian function: (< , > means dot product) is
givenby
H (t , x, , u ) f (t , x, u ), (t , x, u )

Thecostateequationis
H x f xT (t , x, u ) x (t , x, u )

(4)
(5)

Sincewearemainlyinterestedinapplication,westate
the necessary conditions of optimality without proof.
For details see (Ahmed 1988, Kirk 2004, Benson and
Franklin2008).Theyaresummarizedasfollows:
In order for a (control/decision) policy uo over a plan
periodI[0,tf]tobeoptimal,itisnecessarytohavea
costate corresponding to the policy uo and the
associatedstateprocessxosatisfyingtheinequality(7),
given below and the system of equations (8) and (9)
subjecttotheboundaryconditions(6):

t0

tf

H (t , x o (t ), o (t ), o )dt

H (t , x

(t ), o (t ), )dt (14)

t0

AlltheaboveequationswillbeusedintheNumerical
Algorithmsectiontofindtheunknownparametersof
the system. Once these parameters have been
determined,wehaveadynamicmodelfortheenergy
production system. The planner can then use this

x o H (t , x o , o (t ), u o (t )) f (t , x o (t ), u o (t )) (8)

f xT (t , xo (t ), o (t ), u o (t )) x (t , x o (t ), u o (t ))

(10)

tf

o H x (t , xo (t ), o (t ), u o (t ))

o f xT (t , x o , o ) o ( y (t ) x o (t )), (t f ) 0 (13)

H (t , x (t ), (t ), u (t )) H (t , x (t ), (t ), u ) (7)
o

x(t , ) y (t ) dt

TheHamiltonianfunctionforthesystemidentification
isgivenbythefollowingexpression
1
2
H (t , x, , ) f (t , x, ), x(t , ) y (t ) (11)
2
where < , > means dot product. The corresponding
necessary conditions of optimality can be stated as
follows:Fortobethebestparameter,itisnecessary
that the triple {,xo,} satisfies the following set of
equationsandinequality
x f (t , xo , o ), xo (0) x0
(12)

NecessaryConditionsofOptimality

Historical data for all eight power sources are

available and can be used for y(t), t I in the
expression (10). As stated in (Ahmed, 1988), system
identificationusingoptimalcontroltheoryapproachis
possible through the necessary conditions, as
indicatedbelow.

using the state Equation (3), costate Equation (5) and

the Hamiltonian function Equation (4), optimum
decision policy is determined to minimize the cost
functionEquation(2).

tf

where |.| denotes the Euclidean norm in R{8}. The

problem is to determine the parameter o that
minimizes the identification error as defined by
Equation(10).

wherefxistheJacobianmatrixofthevectorfandfxTis
itstranspose.Theboundaryconditionsareasfollows:
(6)
x (0) x0 , (t f ) Tx ( x (t f ))

1
2

(9)

313

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

modelandfindtheoptimalpowergenerationstrategy
foreachofthepowersourcesmentionedabove.

the level of electric power generation and the

corresponding level of production of greenhouse
gasses for each of the sources. This relation is well
known to engineers and it is quite reasonable to
assumethatitisgivenbyanexpressionoftheform:

Energy Production Policies

Once the parameter identification problem is solved,
onecanfocusontheenergyproductionplanning.For
this, different types of cost functions are considered
whichreflecttheshorttermandlongtermsocialneeds.
Forthepurposeofthispaper,threemainproblemsare
considered. The first one aims at reaching the final
desiredgenerationlevelsforeachofthepowersource.
The second one also is concerned with reaching the
goal while keeping the implementation cost to a
minimum. The third problem is concerned with
minimizing the usage of polluting power sources
whilereachingthetotaldemandattheendoftheplan
period.

F ( x) v x , q 1, v 0

(17)

Onlybyexperiment,canengineersdeterminethetwo
parameters{,q}foreachoftheconventionalsources.
It is known that petroleum produces 1.23 times more
CO2 than coal for the same level of electric power
generation (CO2, EIA, 2012). This relationship is
consideredasareferenceinourproblemfornumerical
results. For numerical purpose, we choose q = 2. The
objectivefunctionalforthisproblemisgivenby
tf

J (u )

8
8
1
{ qi ui2 (t ) wi xi2 (t )}dt

2 t i 1
i 1
0

(18)

1
( X d (t f ) xi (t f )) 2
2
i 1

(P1)OptimumDecisionPolicyforMeetingtheTarget
Here the primary objective is to meet the production
goal at the end of the plan period. For this, the cost
functionalisgivenby
1 8
J (u ) ( x(t f )) { i ( xi (t f ) xid )) 2 } (15)
2 i 1

where the first sum represents the implementation

cost and the second sum represents the penalty for
environmental damage due to usage of coal and
petroleum. The third term represents the mismatch
between the total energy demand Xd and the sum of
theenergiesproducedfromeachofthesources.

where {xid} are the desired target levels of generation

fromeachofthesourcesconsideredand{xi(tf)}arethe
actualgenerationlevelsreachedattheendoftheplan
period.Theobjectiveistofindanoptimalpolicyuofor
thesystem(1)thatminimizesthisfunctional.

Numerical Algorithm
Weconsiderthethreeoptimizationproblemsgivenby
Equations (15), (16) and (18). Before considering the
optimization problems, we must identify the system
parameters { ij}. We use gradient technique along
with optimal control theory to determine the fifty six
unknownparameters{ij}basedontheU.Senergy
data.Thentheseparametersarefixedandthemodelis
set. Next, we use the cost functionals (15), (16) and
then(18)todeterminetheoptimumdecisionpolicyfor
integrating the renewable power sources into the
electricity power generation (Miah, Ahmed and
Chowdhury 2012, Ahmed 1988) based on MATLAB
software(Wang,2009).

(P2)OptimumPolicyforMeetingtheTargetwith
ImplementationCost
Given the current levels of power generation from
eachoftheavailablesources,andtheplanperiod,the
objective is to reach a prespecified target level of
generation while keeping the implementation cost as
low as possible. For this, one must use the terminal
cost function including the implementation cost. This
isgivenbythefollowingexpression
tf

J (u )

8
1
1 8
{ qi ui2 (t )}dt { i ( xi (t f ) xid )) 2 } (16)

2 t i 1
2 i 1

MethodologyforSystemIdentification

where{qi,i0}aresuitableweights.Theplannercan
choosetheweightsinthecostfunctionasdesired.

Step1 Subdivide the plan period into N equal

subinterval.

(P3)OptimumPolicyforMeetingtotalDemandwith
ImplementationandEnvironmentalCost

Step2 Integrate the state equation (12) for the plan

period with any assumed ij and the given initial
conditionsatt0.Let{n,xn}denotethevaluesofand
thestatexatnthiteration.

Among the primary power sources chosen in this

article, Coal and Petroleum are the most polluting
ones.Naturalgasandtherenewablesourcesareclean.
For this problem, we need the relationship between

314

Integrate the costate equation (13) backward

intimewiththepair{n,xn}replacing{o,xo}withthe

Step3

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

terminalconditionsattfasgiveninequation(13).This
willgivethecostaten.

optimum energy policy, it is necessary to define the

desiredvaluesforeachofthepowersources.Basedon
the 20% Wind Energy by 2030 scenario released by
(NREL, 2008), the generation level of Wind power
(source) should cover around 20% of the total
electricitygenerationbyyear2030.InthecaseofSolar
powersource(EERE,2008),providing10%ofthetotal
electricity generation by year 2025 is the possible
desired value. Finally according to (C2ES, 2012), 223
gigawattsofnewgeneratingcapacityisplannedtobe
addedtotheelectricpowerindustryoftheU.S.Thisis
based on the increasing population and electricity
demand of the country. Natural Gas power source
stands for 60% of this increased capacity, while Coal
powersourceisresponsiblefor11%.

Step4 Determine the gradient vector using the data

{n,xn,n}asfollows:
tf

gn

t0

tf

H
(t , xn , n , n )dt f T (t , xn , n ) n dt (19)

t
0

Determine if the inequality (14) is satisfied

with{n,xn,n}replacing{o,xo,o}.Stoptheiteration
if it is satisfied (because the triple {n, xn, n} is
optimal).Ifnot,checktheinequality
Step5

tf

H ( xn (t ), n (t ), n

dt

(20)

t0

satisfied. Stop the iteration if it is satisfied. If not,

replacenfornextiterationusingn+1givenby
n 1 n g n

(21)

TABLE1GENERATIONLEVELOFELECTRICPOWERSECTORS(TRILLION
KILOWATTHOURS)HISTORICALDATA(ANNUALENERGYOUTLOOK,

2012)

andcontinuethesearchfromStep2.

Coal
Petroleum
NaturalGas
ConventionalHydropower
Wind
Solar
Geothermal
WoodandBiomass

MethodologyforComputationofOptimumDecision
Policy
Herealsowechoosethegradienttechnique.Thusthe
procedures is similar to that for parameter
identificationwiththefollowingmodifications:

The parameter n is replaced by un(t), t I

which denotes the decision policy at the nth
iteration
Equation(19)isreplacedby
g n (t ) H u ( xn (t ), n (t ), un (t ))

www.ijesci.org

Inequality(14)isreplacedbyequation(23)

H u ( xn (t ), n (t ), un (t ) o( )
2

Year2015
1.562
0.026
1.028
0.29543
0.15097
0.00647
0.01868
0.02128

In this paper, the target (or desired) values are based

onthelatestdatareleasedby(AnnualEnergyOutlook,
2012) for Low Renewable Technology Cost Case
shown in Table 2. However, our procedure can be
used for any desired target. All the data presented
here will be used in the following section for system
identification and then this model is used to generate
theoptimumdecisionpolicy.

(22)

H ( xn (t ), n (t ), un 1 (t ) H ( xn (t ), n (t ), un (t ))

Year2010
1.831
0.034
0.898
0.25532
0.09449
0.00128
0.01567
0.01151

(23)

Numerical Data

TABLE2GENERATIONLEVELOFELECTRICPOWERSECTORS(TRILLION
KILOWATTHOURS)PROJECTIONBASEDONLOWRENEWABLE
TECHNOLOGYCOSTCASE(ANNUALENERGYOUTLOOK,2012)

Increasing energy demand and environmental

concernsmustbeaddressedinawaysothatdemand
is met while impacts on environment is kept to a
minimum.Withthisgoalinmind,wehavechosenthe
available data from U.S. Energy Information
Administration (EIA) website (U.S. EIA). For system
identificationconcernedwiththedeterminationofthe
unknownparametersofthemodel,historicaldatafor
allthechosenenergysourcesaretakenfromthelatest
releaseddataofEIAwebsite(AnnualEnergyOutlook,
2012).ThisisgiveninTable1.

Numerical Results

After determining the unknown parameters of the

systemmodelusingthehistoricaldataofTable1,one
can focus on optimization based on this model. For

The simulation results obtained by the use of

MATLAB software using the numerical algorithm
mentioned in section VI are presented here. The first

Coal
Petroleum
NaturalGas
ConventionalHydropower
Wind
Solar
Geothermal
WoodandBiomass

Year2015
1.562
0.026
1.028
0.29543
0.15097
0.00647
0.01868
0.02128

Year2035
1.780
0.028
1.037
0.32178
0.31055
0.0869
0.05089
0.07841

315

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

part deals with the problem of system identification

using the U.S. energy projection data over six year
period. This gives the unknown parameters {ij}
includingu{ui}.Weusetheidentifiedparameters
{ij} for the second part dealing with optimization.
Here three different scenario are taken into account
givingtheoptimumdecisionpolicyforeachofthem.

(a)

SystemIdentification
Thepurposeofthissectionistodeterminethefiftysix
unknown parameters of the system based on the
historical data of Table 1. This is achieved by
minimizing the (error) functional given by the
expression (10). Minimizing this functional is
equivalent to minimizing the gap between the model
responseandtheactualdataoftheTable1.Theresults
areshowninTable3andTable4.Figure1showsthe
identification error as a function of iteration. After
10000 iterations, the error reduces to almost zero.
Figure2showsasampleofparametersasfunctionof
iteration.Itisclearfromthefigurethattheyconverge
toconstantvalues.

(P1)OptimumDecisionPolicyforMeetingtheTarget
Here the cost functional is given by equation (15) as
reproducedbelow
1 8
(24)
J (u ) ( x(t f )) { i ( xi (t f ) xid )) 2 }
2 i 1
wherethevaluesofiareshowninTable5.
TABLE5ASSUMEDVALUESFORWEIGHTSINEQUATION(24)FOR(P1)

ij

Value

ij

Value

ij

Value

ij

Value

12
21
13
31
14
41
15
51
16
61
17
71
18
81

0.00226
0.00971
0.02771
0.02508
0.00944
0.03931
0.00460
0.14887
0.00089
0.07470
0.00073
0.01529
0.00407
0.18045

23
32
24
42
25
52
26
62
27
72
28
82
34
43

0.00495
0.00109
0.00356
0.00072
0.00046
0.00176
9.28e5
0.00238
1.20e5
0.00038
3.53e5
0.00735
0.00070
0.01928

35
53
36
63
37
73
38
83
45
54
46
64
47
74

0.00676
0.08901
3.55e6
0.03664
4.35e5
0.00750
3.20e5
0.08850
0.00202
0.02075
2.75e5
0.01041
0.00033
0.00213

48
84
56
65
57
75
58
85
67
76
68
86
78
87

0.00024
0.02516
0.00010
0.00385
0.00027
0.00078
0.00093
0.00931
0.00063
1.07e5
0.00046
0.00012
0.00009
0.00054

TABLE4COMPUTEDVALUESOFPARAMETERSAFTER10,000ITERATIONS

ui

Value

ui

Value

ui

Value

ui

Value

u1
u2

0.019
0.194

u3
u4

0.097
0.128

u5
u6

0.418
0.959

u7
u8

0.191
0.701

Value

Value

Value

Value

1
2

8
5

3
4

3
3

5
6

4
20

7
8

5
10

The objective is to find an optimal policy uo for the

system (1) that minimizes this functional. For this
problem initial values are taken from the data shown
in Table 2 for the year 2015. The data for the desired
values are based on the projection of the U.S. Energy
InformationAdministrationpresentedalsoin(Annual
Energy Outlook, 2012) and shown in Table 2 for the
year2035.UsingthemethodologygiveninsectionVI,
theoptimalpolicyiscomputedasshowninFigures3
5.Figure3showstheoptimalpathofgenerationlevel
from each of the considered sources. Figure 4
shows the optimal decision (i.e. generating) policies
corresponding to all major energy sources in the
United States. Figure 5 shows the convergence of the
costfunctionwithincreasingiteration.Itisclearfrom
theFigure3thatthetargetsetbyU.S.EIAismetvery
closely by following the control policy shown in
Figure4.

TABLE3COMPUTEDVALUESOFPARAMETERSAFTER10,000ITERATIONS

(a)

(b)

FIG.3(P1)STATETRAJECTORIESFORTHEPLANPERIODOF20
YEARSFOR(a)COAL,NATURALGAS,HYDRO,WINDAND(b)
PETROLEUM,SOLAR,GEOTHERMAL,WOODANDBIOMASS

FIG.1IDENTIFICATIONERRORAFTER10,000ITERATION

316

(b)

FIG.2CALCULATEDVALUESFOR(a)16,61AND(b)38,83
AFTER10,000ITERATIONS

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

www.ijesci.org

toallmajorenergysourcesintheUnitedStates.
TABLE6ASSUMEDVALUESFORWEIGHTSINEQUATION(25)FOR(P2)

(a)

(b)

(c)

(d)

Value

Value

qi

Value

qi

Value

1
2
3
4

8
5
3
3

5
6
7
8

4
20
5
10

q1
q2
q3
q4

10
5
3
3

q5
q6
q7
q8

1
0.59
1.8
0.68

(a)(b)
FIG.6(P2)STATETRAJECTORIESFORTHEPLANPERIODOF20
YEARSFOR(a)COAL,NATURALGAS,HYDRO,WINDAND(b)
PETROLEUM,SOLAR,GEOTHERMAL,WOODANDBIOMASS

FIG.4(P1)DECISIONPOLICYFORTHEPLANPERIODOF20
YEARSFOR(a)COALANDNATURALGAS,(b)PETROLEUM
ANDGEOTHERMAL,(c)HYDROANDSOLAR,(d)WINDAND
WOODANDBIOMASS

(a)

(b)

(c)

(d)

FIG.5(P1)MINIMIZEDCOSTFUNCTIONAFTER2,000
ITERATIONS

(P2)OptimumPolicyforMeetingtheTargetwith
ImplementationCost
Intheprevioussubsection,itisassumedthatthecost
of implementation of the control policy is negligible.
Here we remove this assumption and introduce the
controlimplementationcost.Theimplementationcost
may involve both capital cost and maintenance cost.
The cost functional for this problem is given by the
expression(16)asreproducedbelow:

FIG.7(P2)DECISIONPOLICYFORTHEPLANPERIODOF20
YEARSFOR(a)COALANDNATURALGAS,(b)PETROLEUM
ANDGEOTHERMAL,(c)HYDROANDSOLAR,(d)WINDAND
WOODANDBIOMASS

tf

8
1
1 8
J (u ) { qi ui2 (t )}dt { i ( xi (t f ) xid )) 2 } (25)
2 t i 1
2 i 1
0

where {qi 0, i = 1 8} are suitable weights

representing the unit cost of control policies. It is
generallyexpectedthatthevaluesfortheseweightsare
dependent on the source type and may be
comparativelylargerforcertainsourcesthanothers.The
valuesof{qi}and{i}areshowninTable6.Againusing
the methodology presented in section VI, we find the
optimaldecisionpolicyasshowninFigures68.Figure6
showsthepathofoptimalgenerationlevels(asfunction
oftime)foreachofthesourcesandFigure7showsthe
optimaldecision(i.e.generating)policiescorresponding

FIG.8(P2)MINIMIZEDCOSTFUNCTIONAFTER2,000
ITERATIONS

(P3)OptimumPolicyforMeetingTotalDemandwith
ImplementationandEnvironmentalCost
The objective functional for this problem is given by
equation (18) which is reproduced below for easy

317

www.ijesci.orgInternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

reference
tf

8
8
1
J (u ) { qi ui2 (t ) wi xi2 (t )}dt
2 t i 1
i 1
0

(26)

8
1
( X d (t f ) xi (t f )) 2
2
i 1

The weights chosen are given in Table 7 and the

results are shown in Figures 911.Figure 9 shows the
level of generation from each of the sources as
functionsoftime.ItisclearfromtheFigure9thatdue
to introduction of environmental cost, the usage of
polluting sources has been substantially cut down
while production from clean sources such as wind,
solarandgeothermal,biomasshasincreasedsufficiently
to meet the demand. Figure 10 shows the optimal
decision (i.e. generating) policies corresponding to all
major energy sources in US. Convergence of the cost
functionisshowninFigure11(a).

(c)

FIG.10(P3)DECISIONPOLICYFORTHEPLANPERIODOF20
YEARSFOR(a)COALANDNATURALGAS,(b)PETROLEUM
ANDGEOTHERMAL,(c)HYDROANDSOLAR,(d)WINDAND
WOODANDBIOMASS

Figure 11(b) shows the total electricity generation

duringtheplanperiod.Electricitydemandintheyear
2035, as projected by U.S. EIA (Annual Energy
Outlook, 2012), is 4.7 TKWh while according to our
numerical example it is 5.2045 Trillion kWh.
According to this figure, it is observed that the
generationlevelreachedattheendoftheplanperiod
exceedstheestimatedtargetonlyby6.15percent.

(a)

qi

Value

qi

Value

Value

10
5
3
3

q5
q6
q7
q8

1
0.59
1.8
0.68

1
2

5
5*1.23
10

Conclusion
The optimum decision policies based on the official
released data from the U.S. Energy Information
Administration official website for eight main power
sources such as Coal, Petroleum, Natural Gas,
Hydropower, Wind, Wood and Biomass, Solar and
Geothermalarepresentedinthispaper.
A dynamic model representing the level of electricity
generation from each of the selected power sources
and the (economic and availability) interactions
between them is considered. From optimal control
theory,Pontryaginminimumprincipleischosenasthe
appropriate method of optimization. Based on the
available historical data, all unknown parameters of
themodeledsystemareidentified.Fortheplanperiod
of 20 years (20152035) the optimum decision policies
aredeterminedforthreedifferentproblems.

(a)(b)
FIG.9(P3)STATETRAJECTORIESFORTHEPLANPERIODOF20
YEARSFOR(a)COAL,NATURALGAS,HYDRO,WINDAND(b)
PETROLEUM,SOLAR,GEOTHERMAL,WOODANDBIOMASS

Inthefirstproblem(P1),optimalpolicyminimizesthe
gap between the target and the actual levels of
generation for all the eight power sources. In the
second problem (P2), optimum policy minimizes the
gap between the target and the actual levels of
generation while keeping the implementation cost as
low as possible. In the third problem (P3), the
optimumpolicytriestosatisfythetotaldemandatthe
finaltimewhilekeepingtheimplementationcostand
environmentaldamageaslowaspossible.

(a)

318

(b)

FIG.11(P3)(a)MINIMIZEDCOSTFUNCTIONAFTER3,000
ITERATIONS,(b)TOTALLEVELOFINSTALLEDELECTRICITY
GENERATIONDURINGTHEPLANPERIODOF20YEARS

TABLE7ASSUMEDVALUESFORWEIGHTSINEQUATION(26)FOR(P3)

q1
q2
q3
q4

(d)

(b)

InternationalJournalofEnergyScience(IJES)Volume3Issue5,October2013
DOI:10.14355/ijes.2013.0305.03

The procedure used in this article for system

identification and optimum decision is completely
general and the planner can easily adapt this
(optimization) technique to consider any other factor
suchasinvestmentcost,demandsasfunctionoftime,
andothereconomicfactorsbyjustmodifyingthecost
function and applying the same procedure. The
dynamic model used here can be applied to any
economic system consisting of cooperating and
competing agents in a closed economy. However, the
identification and the optimization procedure
presented here are applied to any economic sector
provided that the dynamic model is chosen
appropriately.

www.ijesci.org

http://www.eia.gov/environment/emissions/carbon/
EERE, U.S. Department of Energy, Energy Efficiency and
Renewable Energy, EERE Network News, Study: Solar
Power Could Provide 10% of U.S. Electricity by 2025,
June2008,http://apps1.eere.energy.gov/news/news_detail.
cfm/news_id=11835
Goel N.S., Maitra S.C., Montroll E.W.,Onthe Volterra and
the nonlinear models of interacting populations,
AcademicPressInc.,1971
KirkD.E.,Optimalcontroltheory,Anintroduction,Dover
publications,Inc.,Mineola,NewYork,2004
Miah M.S.,Ahmed N.U., Chowdhury M.,Optimum policy
for integration of renewable energy sources into the

REFERENCES

powergenerationsystem,EnergyEconomics34,558567,

Ahmed N.U., Elements of Finite Dimensional System and

2012

Control Theory, Pitman Monographs and Surveys in

Moss D.L., Kwoka J.E., Competition policy and the

Pure and Applied Mathematics, vol.37. Longman

transition to a lowcarbon efficient electricity industry,

Scientific and Technical with John Wiley, London, New

TheElectricityJournal23,Issue7,615,2010

York,1988

NREL, U.S. Department of Energy, Energy Efficiency and

Annual Energy Outlook 2012 with projections to 2035, U.S.

Renewable Energy, 20% Wind Energy by 2030,

EnergyInformationAdministration,http://www.eia.gov/

IncreasingWindEnergysContributiontoU.S.Electricity

forecasts/aeo/pdf/0383(2012).pdf

Supply, July 2008, http://www.nrel.gov/docs/fy08osti/

Benson S.M., Franklin M., Sustainability and energy

41869.pdf

conversions,MRSBulletin33,2008

Shafiee S., Topal E., When will fossil fuel reserves be

C2ES, Center for Climate and Energy Solutions, Natural

diminished?,EnergyPolicy37,181189,2009

Gas in the U.S. Electric Power Sector, May 2012,

Steinberger C.J., What future for energy and climate?,

http://www.c2es.org/docUploads/naturalgaselectric

PresentationinBadKissingen,Germany,July2009

powersector.pdf

U.S. EIA, U.S. Energy Information Administration Official

CO2, EIA, U.S. Energy Information Administration,

Website,www.eia.gov

Environment Analysis and Projections, U.S. Energy

WangX.,SolvingoptimalcontrolproblemswithMATLAB

Related Carbon Dioxide Emissions, 2011, August 2012,

Indirectmethods,November2009

319