Você está na página 1de 6
Probabilistic a.c. load Tow RN. Allan, M.Se.Tech.,Ph.D.,C.Eng., MLE. Indexing terms: Power-station load Abstract ‘The paper extends the presently available techniques for evaluating the load flow problem probabilistically. It presents two possible formulations of the problem that mit the probabllity-density curves ofthe angles, voltages, injected reactive powers and active and reactive power flows to be computed. Owing to the complexity of the problem, the load-flow equations were linearised. The errors introduced by this technique together with the results of atypical analysis are discussed, of symbols By = imaginary part of element ik of admittance matrix imaginary part of shunt admittance at node i half of susceptance of line / real part of element ik of admittance matrix. number of nodes number of PQ nodes number of PV nodes including slack injected active power at node / active power flow in line /— balance of power at slack node injected reactive power at node i ‘= reactive power flow in line ik injected reactive power by shunt element at node i twansformer tap ratio voltage magnitude at node i reactance of line fk angle at node i referred to slack node difference in angles between nodes / and & expected value standard deviation ‘Subscripts and superscripts § = generation quantity ij = node numbers T= load quantity sh = shunt element = represents inverted matrix and its elements 1 Introduetion Recently a number of papers have been published that have modelled the loadlow problem probabilstically instead of by the Conventional deterministic method. The main reason for this is that the input data of any powersystem load-flow study s known to vary Statistically nd the attempt has been to model these statistical varie ations in the input data. Some authors®* have considered these vat ations tobe normally distributed and hence the output information tobe normally distebuted also. These authors have produced an have restricted their nil forma lation to an equivalent d.. load low but in the process to model the System variable by realistic probability density curves, Using math- tinaticl convolution techniques, the power fiows canbe defined by their probabity-density curves without assuming them to be nor. mally distributed seplthe present paper, the previous formulation!" has been ex- {ended so that ana load flow canbe modeled such that angles lags, active and reactive power flows, reactive power losses and Injected reactive power can be evaluated. 2 Formulation of the load-flow problem the ngs eteinistic load flow study, the known quantities ae {Be injected active powers (at al busbar (PQ and PV) except the ck, the injected eastive powers (Q,) tal oad (PQ) busbars and ‘Rt 2662 trace 3st October 1975 adn coved form 8 Feary Ore Be Aland We. at Shahr are with the Department of Electrol ‘Ranting & cron, Unerty of Marcher tute of Stece & Meloy, Sacre, Manchester 00100, Engrs the voltage magnitude (V;) at all generator (PV) busbars. In general, the load-flow problem can be formulated as B= Bi... Ope Vi =a) o = M(B Byy Me Ty) @ where i= 1,.. 57. From these quantities and equations the system can be solved quickly, efficiently and precisely using one of a number of determin istic load-flow techniques. ‘The problem with these techniques is that, although very precise, they are as inaccurate a the input data, Any variation of this input data can cause very significant changes inthe load-flow solution. To overcome this difficulty, the problem can be modelled probabil- istically"$ ‘The known quantities ina probabilistic study are the same as siven above, However, the quantities F and Q, ate now defined by probability-density curves although the voltages Vj at PV busbars are Still known exactly. A nomber of difficulties are encountered in solving this problem probabilistically since the quantities are no longer single values but are instead, density Functions. Briefly these difficulties are due to (@)8 and V are not avallable in terms of Band Qs (6) the functions and h are nonlinear (©) the random variables P,Q, and V are not necesssily indepen- dent ‘To overcome these difficulties, the problem has been linearised and the random vasables P and Q have been assured to be indepen- dent, Convolution techniques can then be used to deduce the density functions of the unknown quantities. Presently, two linearisation formulations have been investigated. These are based on the pre= viously published" techniques and form the bass ofthis paper. 3 Probabilistic formulations 3.1 Load-flow equations Although well known, itis useful for explaining the prob: abilisti formulation and assumptions made, to first consider the ‘more detailed form of the load-flow equations, These are A= HE IaCGn e080 + By sind) ® 01 = HE MCG sin, By co) o (where 71,50) Pa = ~taGinV?? + VV(Gin C096 + Bn Sv) 0 Oy = Subin¥? ~BiVE+ WV AGn 80H — By cst) (6) Qa) = VP ay io 3.2 Angles and active powers In both of the present formulations, the angles and active power flows were deduced in the same way and with the same assumptions as published previously.""* Briefly these assumptions Ves Ve = Ipu. O because line resistance assumed zero Gn Using these assumptions, it can be shown that (see Appendix 8.1) 8) 0) and (19) In both formulations, asin previous publicstions, eqns. 8-10 were used to obtain the angles, active power flows and balance of power at the slack (reference) busbar. 33. Voltages and reactive powers As discussed in Section 2, it is necessary to linearse the load-flow problem in order to compute the density functions of the voltage and reactive powers, One assumption in the present formu lations was that the active and reactive powers could be considered as, decoupled. In addition, linearisations similar to those assumed for the 4.c. probabilistic load flow were made, Two such methods of linear isations were made which are best named as formulation I and formu lation 2. (@ Formulation 1 To deduce the voltages and injected reactive powers at gener- ation nodes, qn. 4 was linearised by assuming the voltage Vj = Ip. ga. 4 therefore becomes a= 3 Ante ay where Ain = Gin sinb yy ~ By 6050 Au ~ Bu where Rim Xin Rat ta? SP BP ae As shown in Appendix 8, the voltages at al load busbars and the in jgeted reactive powers atl geneation nodes can then be deduced From eam, 11 "To dee the reactive power flows and the reactive power con- tributed by the dunt element, oqas.6 and 7 were lneased by aig n= Gn = Vand VV = Ni Eqn. 6 then becomes Oa, = O3n% + Ane | where 2) wate | ca and eqn. 7 becomes Quen) = YBieny a3) ‘As shown in Appendix 8, the resctive powers can be deduced from eqns. 12 and 13. ©) Formulation 2 In this formulation, approximations were made from the knowledge that the voltages in a power system are close to unity. It was therefore assumed that ya 1tsy y= 18h, ‘where V; and 8V are small deviations in Vj and Vy from unity. Therefore We +5) 0+6K) = 1461;+ 5% neglecting SVM Vit Wenl «sy ve = 2-1 (as) Using eqns. 14 and 15, eqns. 4 6 and 7 ean be linearised from which, a shown in Appendix 83, the voltages and reactive powers can be deduced Both ofthe above formulations transform the nonlinear equations into equations consisting of the sum or difference of independent random variables. Consequently, mathematical convolution tech niques canbe applied ina straightforward? manner to obtain the density functions ofthe unknown quantities. 4 Analysis of a typical system 41 Test system ‘The techniques described inthe previous sections have been applied toa wide range of systems from a small 6 busbar tes system to the 57 busbar IEEE test system. Brief details ofthese systems are shown in Table 1 which includes the sine of the systems, the typeof input distributions used and the executions times for formulation 1 using CDC 7600, The executions times for formulation 2 were almost identical. Its evident from Table | that the execution time increases asthe system size is increased and asthe numberof discon tinuous (binomial and discrete) distributions is increased. This wend is that expected, since the numberof convolutions required with these distributions is greater than that required for normal distri butions To illustrate the application and results, this paper considers in more detail the analysis ofthe 14-busbar, 20dine IEEE test system Similar results were obtained forall the other cases though the details ofthe results differed depending on the type and numberof input Alistributions. The data specified by the IEEE for this system is deter minisic as shown in Table 2. However, the specified data for this system does not include an active power forthe slack busbar. Tere- fore, this was assumed to consist of 10 units each rated at 25MW with a forced outage rate of 0.08. Consequently from the binomial dis bution, the expected value 10x (1-008) x 25 = 230MW # To compare the probabilistic evaluations with those that can be com: puted conventionally, the 14-busbar system using the nodal dats shown in Table 2 and the line data shown in Table 3 was analysed using a conventional deterministic ac, load-flow program, Since the data specified for the IEEE test system is deterministic, the probabilistic nature of the input quantities had to be devised. This data is shown in Table 4. In all cass, the deterministic values were assumed to be expected values. All the busbar loads (indicated by the negative values in Table 4), with the exception of that at busbar 9, were assumed to be normally distributed and a standard deviation was therefore associated with the expected values, For the Toad at busbar 9, a discrete load characteristic was assumed such that the expected value was equal to the deterministic value shown in Table 1. The only other source of generations (busbar 2), indicated by 8 positive quantity in Table 3, was represented by 2 units each having a rating of 22MW with a forced outage rate of 0-09; this giving an expected value equal to the deterministic value shown in Table 1 42 Results Using the techniques discussed in the previous sections it ‘was found that the expected value of the balance of power at the slack busbar 1 was 219-0 MW with a standard deviation of 1627 MW. ‘This compares with a value of 232:39 MW using conventional tech: niques. Also, the expected values and standard deviations ofthe busbar angles, voltages and injected reactive-powers using formu: lations 1 and 2 are shown in Table S and some of the active power flows and reactive power flows are shown in Table 6. In some casts the expected values are compared with those obtained using a conver tional deterministic method. From Tables 5 and 6 it can be seen that the expected values of angles, voltages, active and reactive power flows and-injected reactive powers are similar in magnitude to the values obtained from the deterministic analysis. Te errors that exist are clearly expected how’ ever because of the linearisation process used in the probabilistic ‘model ofthe problem, These errors can be considered tolerable how ever being in mind the significantly increased information that can B¢ ‘gained from a probabilistic analysis. Also the error can be minimised a5 will be discussed later. pre Hee vet 193 An «INE 1976 Number of| Input distributions wv ste Resctve Execution tine system | Bustos | gh, | tines | shane satin ac Baceuton s vt 6 > a 3 binomiais1 | binomial 160 normals sora vee 5 20 1 binomials 146 forma 1eRE “ 5 2 ' Mommie] dieretet 296 disreted sormals? norato rest n ‘ Pa 4 fomaiat | noomats18 oa Tet n 6 % 0 nommale2é | norma 033 bee Ff ‘ a 2 rormaisa9 | nonmaleaa O70 ex Fy : wo 6 formats? | nomalse oe i fa i 0 3 onmaiese | nove 30 168 Tle? distibutions, yet the computed ets diffe sigan fom nor DETERMINISTIC DATA FOR IEEE 14BUSBAR SYSTEM mt ob it al ie 1 ha hose she bility density cores of dost iene shape to those shown Busbar__] voitae | Active powers tad active ig Figs 1-4 were obtained using formulation 2 except that because Number_| Type Pe Generation | Load the expected values obtained from the two formulations were slightly. 1 [atsce | 0600 | mw | sw 36 2 fev} rosso | 4ooe | arze| 1270 2 | ev | 10100 o420] 1900, a | te aa] “390 5 | 80 60] iso 6 | bv] 10700 0] 730g 7 | ro é & | Pv | 10900 g 5 | Po rs0| 1660 44 wo | Po 900/580 = no | 29 bso) ts | Pe aio] to a | 20 1350] 580 | 29 aso] S00 d bles LINE Dsta FoR IEE nuspan syst BE ee Busbar Resistance | Reactance | Susceptance | Transformer ‘active power flow, MW Sag [Recvng] Pe | Pe | Pu pig a , som [asi | acne} a ‘Pbabity denaty carve for ci power flow inline S-6 1 oosaty | o2mos | Goasso Tables i troass | apr | onni90 PROBARILSTIC DATA USED 2 Sosert | turn | tote 2 005695 | 017388 | 0.01700 3 Normal trations 3 oo67a1 | 0.17103 | 001730 Buster | Active power Reactive power 4 0.01335 | o042tt 0-00640 Nember [5 Voltage 4 - 020912 - 22 tart ype o cs 2 4 - | osseia - oH pa | MW] | MVAR % s =, oasaoa - ~68 2 [ev] ross | 2124] 9.00 |~1270 92 6 9.09498 | 019890 > 3 py | taro | —94.20 | 10:00 |~19.00 105 6 012291 | 0-25581 - 4 PO - 47-80 | 11-00 390 97 s 0-06615 | 013027 - s fro] — =160} 5.00 | 160 50 , | Sins | & J ev] sor | “ita | soo | Binomial ane aisorete powers ae increased fora gen system size the =292| 920] -1r0 30 Computed density eure deviate more from anormal dstabuton In ous | 03s) 96 os the present example, most input quantities were defined by normal saa PROC. IB, Vol. 123, No. 6, JUNE 1976 Table 5 ANGLES, VOLTAGES AND INJECTED REACTIVE FOWERS, ‘nals Voltas | Formulation 1 Formulation 2 busbar # a | Deter ee _ _ | H @ _[ we ter oes Pa oa |p| Pa roar | 1088 0-66 vor 10156 ob0s 10188 | 000s "6 9.09 386 ros? | Longs 000s 10166 | 000s 13:36 1391 0-96 0613 | 10593 | 0-0026 10603 99-0027 1493 1569 116 0557 10526 | 0-005! | 0s42 0052 =15-09 =1557 rio 10508 0469 0.0083 1oago |b boa4 =ia79 a6 097 10568 oso 0.0023 hoses | 00025 =1507 =1557 088 hoss2 10526 8.0005 10523 | 00008 =i545 =is14 | 090 1.0503 roaeo 0.0010 1oase | 0010 = 1603 =ii9_ | 6 103s ozs? 0.0038 10292__| O03 Injected reactive power Duan Formulation | Formation? [ae oe | # 2 War Nivar_ | MVar War NAT 1 oo oo “G31 “ni | os Aisio | 023 2 498 oa | 3610 3524 | 075 3593 | 075 3 = 092 | 129 a4 | ose a7 | 4s ri =422 | O84 $68 763 181 50s | Lee s | 1336 | 996 078 ni iar | 68s | tse Table 6 ACTIVE AND REACTIVE FOWER FLOWS “Astive power flows T ‘Reactive power flows tine | bewm he | bem Formulation “Fommalation? | | # 7 | * | © MW MW ; Mwy MVAr MVAT MVAr MVAr | MVAr_ | ase78 ures | Tae | 1893 —18-08 * 2040 : 74-08 ais | 438 562 ses | 023 529 | 029 nit 7001 320 as 439 : 384 : 35°30 | 5515 3-20 — 0-40 0.09 0-46 — 067 046 410? 4097 23 261 an | 029 239 | ons } teaa | 249 fe | — 538 jor | ose <613 | ae 6134 6175 455 eit 163 | 075 162s | O78 175 761 oat 2st aa | G16 237 | 16 45 1725 18 724 | 751 | O6 v6 | Gee 20s | 2836 333 $16 gsr | 228 608 | 34 a0 | -323 iaa_| = 166 =139_| 091 ass | O56 * Te sundard dovntons in hes es reo sine the vlage a both eds f then was seid 2% . 20 2 Boel 8 § 27 3 Jie _ - 6 1G Ts Te i votage. pu I 4 ol eau Fi.2 9 ib it @ 3 Probability density curve for voltage at busbar 4 different, the corresponding curves were shifted slight relative to teach other. This similarity in shape is confirmed by the results shown in Table 5 and 6. Compating the standard deviations given by the two formulations shows that, in practically all cases, the results are vir tually identical; in a few cases a very small difference exists Both of these aspects are interesting features since they indicate that the computed density curves and standard deviations are reason: ably precise although small erors exist in the computed expected values. Because ofthis, the errors in expected values may be overcome by shifting the computed density curve, without changing its shape, ‘until the expected value of it coincides with the value obtained from a conventional deterministic analysis. This newly positioned curve would then represent a reasonably precise disteibution for the quantity being computed. Furthermore, the computational efficiency could be improved in such an exercise by using the expected values obtained feactive power flow MYA Fig. 3 Probability density curve for reactive power flow inline 7-4 from either formulation as the starting values for the conventional iterative analysis. 5 Conclusions This paper has shown how the expected values, standard deviations and probability density eurves of angles, voltages, active power lows, zeactive power flows and injected reactive powers can computed using linearised versions of the ac, load-flow equations. This permits the previously published techniques to be extended 4 that the model more truly reflects the needs of a power-system Joad-flow study. The small errors that occur can be compensated fo" (by simply Sucng the computed density curve until its expected value coincides with the value obtained from a conventional deterministic analysis, ensity function x 102 ee a2 injectected reactive power Mir ! : Fig. 4 Probability density curve for injected reactive power at busbar 3 From these density curves, it sa simple exercise to deduce confi- dence limits and the probability of any quantity being greater than or Jess than a certain predetermined value. The importance ofthis in power-system studies cannot be overstressed and forms a vital part in the analysis of data that is known to vary in practice or is in error due to forecasting problems. 6 Acknowledgments The authors ate indebted to Prof. C.B. Cooper for providing the facilities necessary to conduct these studies. Mr. Al-Shakarchi is also indebted to the Iraqi Ministry of Higher Education and Scientific Research for providing him with a scholarship, 7 References 1 BORKOWSKA, B.: Probabilistic oad flow’ JEEE Tront, 1974, PAS93, pp. 152.759 2 ALLAN, RN, BORKOWSKA, B. and GRIGG, Cit: ‘Probabilistic analysis ot power flow.!, Poe IEF, 1944, 121, pp. 1881-1356 3 ALLAN, RIN, GRIGG, Ci, and AL-SHAKARCH, M.R.G.: ‘Numerical techniques in probabilistic load flow problems’ Jn J. Numer. Methods Eng. (tobe pabished) 4 DOPAZO, .F., KLITIN, A. 2nd SASSON, A.M. ‘Stochastic load Hows’, IEEE Trans, 1915, PAS94, pp. 299309, £ ABOYTES, i, and CORY, B1‘An alternative formulation ofthe stochastic load ow method’ Poceadings ofthe PICA Conference, New Orleans 1975, 6 ALLAN, RIN. and GRIGG, Ci: Discussion contibuton to Reference 4 8 Appendixes 8.1 Angles and active powers From the assumptions given in Section 3.2, eqn. 3 becomes aS Ex oy ha” rin matrix form Pace where =staneg= & 1 Ge = Fi antGum Se ‘which the slack busbar row and column are deleted Invertng giv oP 6S Gh Ge nt) (18) Similarly, from eqn. 5 HL G~ Gon an in which if node fis slack, Gy = 8.2 Formulation 1 Writing eqn. 11 in matsix form gives Q=av partitioning into the load and generation quantities gives lw el fe be 0 where yas my elements nd Q, has my elements. From eqn. 18 Q=MVu+LY, y= eqn where H=-Ly writing explicitly gives Yen = 3 taQan +3 teh C= anom) 09) so fom eqn. 18 aN KAry subttutingfr¥, gies = DOHE whee D = Nita E = DIY, sring exit aes Qa = J Dua + = team) (20) from eps 12 4nd 19 On = F ently + Andru +3, ; indy + Aig) IPiand kare load busbas en Oe = J Atl Qn +S Aut + aah if is «generation busbar mo oN Om =F cibloQhey +E ean) + Aine if eis a generation busbar Similar equations can be derived for Qy, and the reactive power doses, in On Finally, from eqos. 13 and 19 Geers = Bass Hr + Prey Y Maly (2) if is load busbar 83 Formulation 2 ‘Substituting eqn. 14 into eqn. 4 EU +K- Day where Ary is defined in Section 3.3 Thus, =H @3) where Ay = Ay fork ei Ais Wad Aw In matix form Q=A'V-w ‘which when partitioned into load and generation quantities gives, wil lh o (| — ey 10 Yel [Mel Using the same technique as described for formulation! gives, mm Yoo = & Min +B Mat om 5) Oe = S Diy +8 fhm (26) fi where H=WA-L% D = NM El =D'H'+I'V,—W, Similarly, substituting eqns. 14 and 15 into 6 gives Qin = Bini + Ain Va +m @n where Bin = 20m, + Ay Te = ~~ Aa ‘Substituting eqn. 25 into eqn. 27 gives a similar set of equations for (Qu. Ona and line losses as that given by eqn. 21 in formulation | above. Finally, substituting eqn. 15 into eqn. 7 gives Qigany = 2Bigany Vi Bicony (28) from which an equation similar in form to eqn. 22 ean be deduced.

Você também pode gostar