You are on page 1of 5

Aspects of using the sensitivity theory in energetically optimization problems


A. Hazi 1 , Gh. Hazi 2 1 University of Bacau 2 C.N. TELETRANS S.A.Bacau Agency

For different technical decisions taken practically it is useful and necessary to know manner how different parameters influence decision. In the optimization problems, it is interesting the sensitivity of the objective function of the process control parameters. This is given by the gradient of the objective function reported to the control parameters. For the optimization problems with constrain, this is given by the gradient reduced. Pseudo optimization is the optimization of the objective function based on the sensitivity theory modifying only some control parameters. For the optimization problems with constrain, it is interesting sensibility of constrains touched reported to the control parameters, too. This is necessary because one control variable must be transformed into one dependent variable when it is touched one constrain. The dependent variables values result from solving e!uation system given by the e!uality constrains.

(! z 1 , z 2 ,.., z%&% = min ' , %&% = $ # " f ai ! z 1 , z 2 ,.., z%&% = 0, i = 1,2,.., " f li ! z 1 , z 2 ,.., z%&% 0, i = 1,2,.., L

&'( where z1,z2,...,zT T are problem variables, fai, i!1,2,..," are e!uality type constrains &active constrains( that involve a number " of dependent variable, and f#i, i!1,2,..,L are ine!uality type constrains &free constrains(. $ is number of independent variables &control variables(. If variables z1,z2,...,zT T are divide in independent control variables %1,%2,..,%$ and dependent variables y1,y2,..,y", then sensitivity of the objective function reported to the control variables is given by the gradient reduced of the objective function )*+%
(r [*$ ] = = ) ( f a f a ( = ) ) % y % y


1. ANALITICAL RELATIONS OF THE SENSITIVITY 1.1. Sensitivity ! t"e #$e%tive !&n%ti n

In the damages situations, when process parameters are out of the admissible limits, must be intervene rapidly for secure process stability. It must be chose that control variable which, modifying, has a ma"imum efficiency, so straightening rapidly. #nowledge this control variable can be realized on base sensitivity theory. The e"pert systems use massive the sensitivity theory. $sing this theory is also due to low precision of input date from process. In this case, a high precision of calculus is useless. It is consider a general form of one optimization problem%

where &r is the objective function reduced to the independent variables. The sensitivities can be also calculated directly by numerical derivate
(r = ) i = ( ! )' , ) * ,.., ) i + ) i ,.., ) $ , y' , y * ,.., y " ) i

, , ( ! )' , ) * ,.., ) i ,.., ) $ , y' , y* ,.., y , " ) i

i = ',* ,.., $

&-( where variables y'1 , y'2 ,..., y'" are determined by solving e!uations system given by e!uality type constrains with independent variables unmodified. y1, y2 ,..., y" are determined by solving e!uations

Aspects of using the sensitivity theory in energetically optimization problems system given by e!uality type constrains with variable %i modified. In the optimization problems not included constrains, the second term dissolves from &*( and sensitivity become e!ual with simple gradient of the objective function It must be precise that sensitivity is a punctual characteristic, valid for an operation point of the process. Its e"tended for the other operation areas must be analyzed
f lr f lr = ) f = l ) ) = f a y

f l [ " ] y

f a )

[ "]


8ased on the relation &7( it can decide introducing an independent variable in the base. It is choose that variable for which constrain attained has ma"im sensitivity.

1.'. Sensitivity ! t"e ine(&)*ity+ty,e % nst-)ins

/ensitivity of the ine!uality type constrains reported to the independent variable &control variables( it is determine deriving constrains e!uality type and ine!uality type. It must be taking account that dependent variables )0+ depend to independent variables )1+ through constrains system%
f a ) f a + y y ) = ,


In this paragraph we present three possibilities for using sensitivity in energetic with e"ample for the optimization problem voltage 9 reactive, )'+. For the numerical results we use o network **,:'', k; with 2< nodes in areas 8acau and 5eamt.

'.1. Pse&. + ,ti/i0)ti n

&2( Practically, optimization is usefully if it is taken with little maneuvers in the installations. =therwise, optimization is difficult, even that through objective function it can be !uantify risks caused by maneuvers. For this reason, recognizing contribution every control action, it can be give up to these lessees efficiently. It is consider an optimization problem voltage 9 reactive, )'+%
(! [. ] , [ ] , [- ] = = 2! [. ] , [ ] , [- ] + 32% i ! 0 p + i i/ l + 3 2L&%i, ! - i, = min % i , ,r imp 2i ! [. ] , [ ] , [- ] 2 = ,, i n 1 e i imp &<( 0 i ! [. ] , [ ] , [- ] 0 i = ,, i c . min . i . ma) , i n i i min 0 0 i ! [. ] , [ ] , [- ] 0 ma) , i / i i min ma) - i, , i , , r i, i, $ i, ! [. ] , [ ] , [- ] $ ma) ! [. ] , [ ] , [- ] , i , , n i,

f a y ) = y

f a )

f lr f l f l y ) = ) + y ) = f f f = l l a ) y y

f a )


f #r % ] it is taking where through notation [ account that derivative it is calculated in the present e!uality type constrains. 4elation &3( gives calculus manner of the sensitivity of ine!uality type constrains reported to the independent variables. Practically, when in an iteration it goes to the optimal solution on the direction (") with a step , it is interesting sensitivity of the constrains reported to this step. 5oting (*') current value of the independent variables, then new value is%

[ + ] = [ + , ] + [ "]


and sensitivity of the constrains reported to step results from &6( and &3(%

where &+(U),(,),(-). is the objective function, )>?+. (U) 9vector of the voltages between phases

Aspects of using the sensitivity theory in energetically optimization problems in the network nodes, )k;+. (,) 9 vector of the voltages phase differences in the network nodes reported to the reference node &e!uilibration node(, )rad+. (-) 9 vector of the plots of the transformers with long transversal control, t 9 its crowd. C/Ti+01i. 9 e"penses, )>?+, realized in the network node i for producing reactive power 01i , )>;@4+. C/L Ti2+3-i2. 9 e"pense for modification plot of the transformer connected between node i and node 2 with amount 3-i2, , )>?h+. /ii41, 0ii41 9 active power, reactive power, imposed in node i by the conditions from this node. the value is given by the algebraically sum between power produced and power consumed for the nodes were there are production groups connected, )>?+, )>;@4+. /i+(U),(,),(-)., 0i+(U),(,),(-). 9 active power, reactive power in node i determine function by the regime date of the network, )>?+, )>;@4+. $i2+(U),(,),(-). current on the side i2 , )@+. n 9 network nodes crowd. e 9 e!uilibration ode. c 9 consumer nodes crowd. 5 9 control voltage nodes crowd. t 9 long transversal control transformers crowd. /ensitivity of the objective function reported to the independent variables is given by the value of the gradient reduced in the initial point. Its components, for a network **,:'', k; with 2< nodes in zone /TA 8acau, are presented in table '.

T)#*e 1. /ensitivity of the objective function reported to the control variables. 5etwork with 2< nodes V)-i)#*e G-).ient -e.&%e. USTE1ARU ,.,363B )>?:k;+ 2GUTINAS ,.,,3** )>?:plot+ 2DUMBRAVA ,.,26*< )>?:plot+ 2STE1ARU ,.,-B77 )>?:plot+ 2BC3SUD ,.,6-3< )>?:plot+

From table ' it is found that objective function has ma"im sensitivity reported to the plot of autotransformer in station 8acCu /ud &the objective function decreases when plot it is increased(. Table * shows results for successive optimization reported to the problem variables, in the succession indicated by the sensitivities and in the succession modified.

In variant ' variable succession is chosen in the sensitivity order from table ' and in variant * the order is modified. It is find that by pseudo optimization final result differ with ,.B'D from value obtained by simultaneous optimization of 3 variables &-.,3B >?(. That notes a decrease of objective function less '' D. Ehanging of optimization succession leads to a bigger value of the objective function &deviation *.,6 D(. That notes a decrease of objective function less *2 D.

T)#*e '. /uccessive optimization reported to the one variable. 5etwork with 2< nodes V)-i)nt 1 St)4e I II III IV V I II III IV V V)-i)#*e / .i!ie. 2BACAU3SUD 8STE1ARU 2DUMBRAVA 2STE1ARU 2GUTINAS 8STE1ARU 2DUMBRAVA 2BACAU3SUD 2STE1ARU 2GUTINAS V)*&e Initi)* Fin)* B '* 7.< *, '* '* '' '' '* '2 7.< *, '* '' B '* '' ', '* O#$e%tive !&n%ti n 5M26 C nt-i#&ti n 576 -.*,,' 3'.7 -.'2<7 *'.B -.'2<7 , -.'2<7 , -.,<6B *6.2 -.*7B7 *,.< -.*773 '.' -.'6*' 37.B -.'33-.2 -.'**, '6.7


'.'. Sensitivity ! % nst-)ins -e, -te. t % nt- * v)-i)#*es

In problem &<( it is used a sensitivity matri", )/+% &B(

0/ . / 4*5 = . c . / $ . /

] [ . c - ]
[ $
- ]

] [ 0/ - ]

Aspects of using the sensitivity theory in energetically optimization problems It was calculated using numerical methods, with relation &-(. /ome values of sensitivities are given in table -. /ensitivity values show how constrain it is modified when it is altered control variable corresponding. For network analyzed it is find zone influence of control voltage. For e"ample node /tejaru '', #; modifies voltages in area Piatra 5eamF 4oman. In this problem, activate constrain 9 reactive power in node /TGH@4$ '', k;, leads to introducing this node voltage in base. ?hen this voltage it is blocked to the ma"im limit, it is introduced in base plot @T 3 26 because reactive power &and voltage( has ma"im sensitivity.

T)#*e 9. /ome values of sensitivities. 5etwork with 2< nodes C nst-)in 8STE1ARU 8STE1ARU 8STE1ARU 8STE1ARU UM?RGINENI UM?RGINENI UM?RGINENI UM?RGINENI IAT@GUTINAS IATBACAU3SUD In.e,en.ent v)-i)#*e 2GUTINAS 2STE1ARU 2BACAU3SUD USTE1ARU;11<= 2GUTINAS 2BAC?U3SUD 2STE1ARU USTE1ARU;11<= 2GUTINAS 2GUTINAS UM MVAR:,* t MVAR:,* t MVAR:,* t MVAR:>V >V:,* t >V:,* t >V:,* t >V:>V A:,* t A:,* t Sensitivity -.,36BB*2<'7 6.7--2723*'< *.-6,3*3'<*' B.<'<B<6B2-B ,.76*<2362-2 ,.3*B-',6*6' ,.,6'762<7-* ,.*,3-<6<72' *-.**B,737<< ',.,2B3767*

In this paper it was also studied how modifying of operating regime influences sensitivity matri" )/+ in the optimization process. For simplify, it was considered initial regime where are modified some variable. It was find that, although regime analyzed is much far off to initial regime &power loss is up 6, D( voltage sensitivity in consumer nodes it is modified, generally, with value to ', D. /ensitivity of site current reported to plots it is modified in large limits &to 7, D(. It is considered that using sensitivities calculated in reference regime for optimization some regimes is satisfactory. That is because in calculus procedure it is interesting, usually, only sensitivity order of different variable reported to control variables and, practically, real regimes are very close to optimal regimes. For optimization problem, where sensitivity it is using only for choosing variables that enter in base and do not for control strictly speaking, modifying of sensitivity is even less important. @t the same time, it must be take account that e"ceeding site current is unusually, practically, in normal regimes.

consider a node ic, where we want increase voltage with value 3Ui, the correction necessary is%

.i =

.i . , # ,/ . ,

.i + - i, , i c i, ,r i,


'.9. V *t)4e % nt- * in .)/)4e -e4i/es &sin4 sensitivity /)t-iA

In case damage, important decrease of network voltage it must be annulled !uickly, )*+. ;oltage sensitivity in node &or in nodes( with low voltage reported to control variable is characteristic which allow taken decisions !uickly. If it is

The control can be realized throw one or many means. It is recommended that used little means. For e"ample, we consider that station **,:'', k; /tejaru, including @T and lines **, k; and '', k;, turn out due to a damage. Gvidently, it must be used sensitivities calculated for this operating scheme or one close by it due to operating scheme is modified significantly. Eontrol possibilities are plots of @T in stations Aumbrava, Iutinas or 8acauJ/ud. In this regime, voltages in some nodes in Piatra 5eamt and 4oman are affected. ?e consider voltage control in node Eiritei '', k;, which has value ',6.* #; in this regime. /ensitivities of this voltage reported to plots of @T are presented in table 2. T)#*e @. ;oltage sensitivities in node Eiritei reported to plots of @T P* t ! AT 2GUTINAS 2DUMBRAVA 2BACAU3SUD

Aspects of using the sensitivity theory in energetically optimization problems Sesitivity 5>V:,* t6 ,.236 '.,B7 ,.-3For problem of ine!uality type constrains control, this method offers the best solution for introducing an independent variable in base. ;alues calculated for sensitivities are valid for a large domain of regimes in case of voltages in consumer nodes. $sing sensitivities for establishing succession in pseudo optimization process leads to the final results better than in case aleatory order. In the damage regimes, when we want to return !uickly of some parameters in the admissible domain, this method leads to the very well results if sensitivities are calculated for regimes close it.

If it is imposed increasing of voltage up ',B #;, it must be modified plot of @T in station Aumbrava &because voltage in this node has ma"im sensitivity( with value%
',B ',6.* - ".896A:A = 6&.7" = '.,B7

&''( ;oltage estimated is ',6.* K - '.,B7 L ',B.2B #;. @fter calculus of regime with plot of @T modified from '* to '3, results voltage value in node Eiritei ',B.33 #;, so very close by the value estimated with sensitivity. /o, for control voltage in damage regime, it is necessary to have sensitivity coefficients calculated for operating scheme close by damage scheme.

1. Hazi, Gh. Consi6erarea caracteristici#or statistico71ro8a8i#istice 9n o1ti4izarea regi4uri#or siste4e#or e#ectroenergetice, Tez: 6e 6octorat, Universitatea Te;nic: <=;.Asac;i>, $a?i, 1.@ABC', 1DDE. 2. Hazi, Gh., Hazi, A., Ba#anFe ?i ca#itatea energiei, E6itura <TEGN$CA7$N& > C;i?in:u, 1.1CAB1CD, 2''H.

$sing sensitivity theory in energetic is a necessity because it must be a rapid response in the optimization problems or in the damage regimes.