FORECASTING ELECTRICAL ENERGY DEMAND OF SRI LANKA: GENETIC ALGORITHMS BASED APPROACH A dissertation submitted to the Department of Electrical Engineering, University of Moratuwa in partial fulfillment of the requirements for the Degree of Master of Engineering by D. Y. T. BAMBARAVANAGE Supervised by: Dr. Lanka Udawatta Department of Electrical Engineering University of Moratuwa, Sri Lanka 2005 85961 Abstract A novel approach for electrical energy demand forecasting (Short term projection) using genetic algorithms is presented. This model is based on genetic algorithms. Possible factors that affect the electrical energy demand of a system have been counted as variables for the model. By subjecting real time past data for 18 years, on each factor, to natural evolution the forecasting model was obtained. Validation of the model has been carried out and results show the effectiveness of the proposed technique. Forecasting is both a science and an art. The need and relevance of forecasting demand for an electric utility has become a much discussed issue in the recent past. This has led to the development of various new tools and methods for forecasting in the last two decades. In the past, straight line extrapolations of historical energy consumption trends served well. However, with the onset of inflation and rapidly rising energy prices, emergence of alternative fuels and technologies (in energy supply and end use), changes in lifestyles, institutional changes etc., it has become very important to use modeling techniques which capture the effect of factors such as price, income, population, technology and other economic, demographic, policy and technological variables. There is an array of methods that are available today for forecasting demand. An appropriate method is chosen based on the nature of the data available and the desired nature and level of detail or forecasting. The proposed methodology is based on Genetic Algorithms, where all possible factors that affect the electrical energy demand of a system are considered. The forecasted electricity demand with this model for the last two years was with more accuracy compared to the Ceylon Electricity Board forecasted demand; i.e. the modal forecasted demand in each year (year 2002 and 2003) was very much closer to the actual data. DECLARATION Ihe \\ ork submitted in this dissertation is the result of my own investigation, cxc.:ept where otherwise stateu. j It ha~ not already been accepted for any degree, and is also not being concurrently submitted for any other degree. 0 0 0 .:P.~J)b_~4. ~-~- 0 0 D. \'. ·1. Ban;b~a·t{age . ~~- !uj "')..,oos Dati·- I endorse the declaration by the candidate. )-s(l' /? ..-s oooooooo ooidooo 0000 ooooo 00 0 I )r. I .anka Udawatta 11 • ':t' :. :..... 'f I . ·'· -- ./.., ACK~O\VLI~DGE:vl ENT First and foremost my sincere thanks go to the Department of Electrical Engineering, l 'nl\erstty of \1orattma for sclectmg me to folio\\ this course of i\lastcr of Enginccnng (in l ~ lectrical Engineering). \~ my supcrYisor and the course coon..linator. thank you vert much Dr. Lanka Udawatta lor the valuable support anJ guidance given me to make this research a sm:ccss. My special thank goes to the former uircctor Dr. T.A. Piyasiri and the deputy registrar \lrs. Vishakha Korale of Institute of l'cchnology. University of \1oratuwa for granting the course fee in doing this postgraduate course. The tn,aluable guidance anJ support I received from Dr. Thilak Siyambalapitiya and Dr RohJn \1unasinghe '' hde writing this thesis IS remembered ''ith thanks. In collecting u:.lta the support I recciH!d from my colleague Mrs. \ldhavi Kudaligama or the Ceylon Electricity Board. officers or the DepartmentA~ Senses and Statistics. and .. \lr. \\ asantha of the Central Bank of Sri L.1nJ...a is highly appreciated with thanks. Thank you very much Mr. A.G. Buddhika Jaya::;ckara, lor the kind co-operatlon'and the help given me throughout the project. ..... ivly special thanks extend to the I lead of the Department of Electrical r,:,nginc,-ering- /., Pro!'. H.Y.R. Perera. all aca<.lemic and notHtc:.~demic staiTmcmbcrs of the departmqnl for the facilities and support rendered 111 nttmcrous ways in doing this research. \' i\1 last but not least, the guidance, encouragement and the support I received from my tkar amma, thalhlha and my husband i\lllil, to make this event a success, arc remembered \\' tth thanks. t'lwrangika Bambara\·anage. l 6~ " VI ..... "" I , ·""'· TABLE OF CONTENT 1-t s·r ()F FIC;UJ{ES .................................................................................... 3 I J IS--1~ O li~ ·1~;\ .Bi . ...#l~S .......... ... ..... ............. ....... ................ ........ ......... ..... ...... .... S ('I l .-\ p·r I~ I{ 1 ............................................................................................... 6 Introduction .................................. .. ................................................................................. (> 1.1 Concept ofGA and e\·olutionary programming .............................................. 6 1.1.1 \Vhat are Genes? ...................................................................................... 8 1.1.2 I. 1.3 1.2 1.3 EYolutionary Computation ...................................................................... <.> Genetic Algorithms ................. ....................... ......... .;. .......................... I 2 Electrical energy demand !orcrJsting .......................................................... 16 Proposed methodology of clectricul energy deman<..l forecasting .................. 18 ('II A P'I' E f{ 2 ................•.............................................•........................... ~ .. 19 Sun ey of Avai lable Electrical Energy Demand rorecasting .:V1etho<..lologics ....... .. ...... 19 2.1 Time Trend Method ............................................................ .. ............. ... ......... I 9 2.2 End-usc method ............................................................................................. 21 2.3 Econometric approach .................................................................................. 22 2..-l Combining econometric and time series modcls ........................................... 23 2.5 Load forecast {or 2001 Generation 1:\pansion planning Studies'' ith Econometric Method . ... . .. . . . . . ....................................................................... 24 2 . .5. I Domestic sector...... . ... . .................................................................... 24 2.5.2 Industrial and Commercial Sector ......................................................... 2.5 2 . .5.3 Other sector ........................................................................................... 25 ( 'l lr\ P'f ER 3 ............................................................................................. 31 ... ~. Proposed Methodology Evolution or Electrical Energy D~1and: Genetic Algorithm b;.~scd ;"\1odel ..................... .............................................................................................. 31 3. I Proposed Methodology ..................................... ... .......................................... 3 I 3.5 Just ification of selecting the considered !~lctors ....... ................................. r .. 35 3.5. I Rainfall data. in catchments arc;.~s ................................................. ~:":':· ..... 36 3.5.2 Domestic consumer account<; .............................................. :.::.: ............ 38 3.5.3 Average US S \'aluc ................... .-:·: ......................................................... 38 ~.5.4 Population, Population growth rate .......................................... : .. ~.~ ... 3<) ..... 3.5.) GOP ....................................................................................................... 40 ~- >" , .. . I . v ~ .. :--..6 [ nee of Elcctnctly ( !)omc::;ttc) ......................................................... ..-:...+I 3.6 Parametric Study .............................................. ...................... : ..................... 42 3.7 Limitations in consiJcring all the possible I"Jciors ..................... : .................. 45 Page I of8 (J·IAP'I' l•: l{ 4 ............................................................................................. 47 Results ........................................................................................................................... 47 4.1 Parameters set in the CiC'nCIJC' J\lgorithm ...................................................... .4 ... 4.2 Results used in the forecasting i\.1odci ....................................................... : .. .4S 4.3 Graphical Representation of C\ olut ion of' the parameters ............................. 50 <~H,\PTER 5 .......................................................................... ... ........ ........ 62 Forecasting \lode! for Electrical E .. ·rc~y Demand of S1 i Lank:.! ................................... 62 5.1 Preparing the forecasting ~lode! ................................................................... 62 "2 . Frror with the forecasted data ....................................................................... ().) S 4 \'alidity of the forecasting \1odcl. ................................................................. 64 5.5 Electricity Demand forecast ................................................... 1 ..................... 65 SJ> Conclusion ........................................ ........... ................. .. ............. .. .... ............ 68 lt~:FEl{ENCES ......................................................................................... 70 ,\I'PENOIX 1 ................................... ... ....................................................... 71 F1tness \<.dues obta ined with Jirtcrent pa rameter scllings .................... .. ....................... 71 ,\J) I)~:NI)IX II ........................................................................................... 85 Data used for the forecast .............................................................................................. 85 --~ " ... .. ,,. I , #,' ~ /~ Page 2 of88 LIST OF FIGURES Frgure 1.1: A Genetic Algorithm cycle 7 rigure 1.2: Frgurc 1.2: Search techniques II Figure I.J. Artificial Intelligence techniques 11 Frgure 1.4: Basic.Evolution Cycle 12 Figure 1.5: Roulette \\'heel selection 15 Figure l.(l: Example of one point crosso,·cr · 15 Figmc I. 7: Process of mutation 16 Frgmc 2. I: l:lcctricity demand forecasted by The CCB in yr. 200 I: ShortAem1 projections 20 Frgurc 2.2: The electrical energy demand of Sri Lanka from year 1985 till year 2002 28 h , nre 2.3: Electrical Energy demand forecasted by The CEB in year 2001: Long term 30 Figure J.l: Representation of genes 32 Frgurc 3.2: Way to obtain the forecasting model 32 Frgmc 3.3: Distribution of fERROR2 \ s. Number of generations 34 hgurc J 4: Factors considered in the forecastrng model 34 Frgurc J.Y Consumption of electrieit} by mam consumer categories of Sri Lanka 35 Frgurc 3.Cl: Consumption of electricity hy main consumer categories of Sri Lanka ,l . as a percentage of the total consumers of that year .!' Figure 3.7: Consumption of electricity by Commcn.:ial, Domestic, lnJustrial and Other consumers in Sri Lanl--a in year 2003. Figure 3.8: Rainfall in catchment areas . ·~ ,. l"rgurc J.lJ: Estim:.~tcd mid year population' s. Year Frgmc 3.10: The variation of Economic Growth and Demand lor Energy in Sri ..... Lanka. .., Frgurc J. 1 1: The ,·ariation of the GDP in L S <) agarnst year Frgurc 4.1: Sample fERROR2 's. 0Jo. or gcncrations-dt::;tnbution with I 000 gcner:nions 35 3(l .., 37 39 40 41 48 Page 3 of SX Figure 4.2: Distribution of best fitness vs. no. of gcncra::ons Figure 4.3: Parameters obtained at the best-lit \'alue F1gmc 4.4: Distribution of Variable I (a I) vs. :\umber of' Generations l·igun.: -L5: Distribution of Variable 2 (pI)' s >-: .nbcr of generations Figure 4.6: Distribution of Variable 3 (a2) \S. >-lumber of generations F1gurc 4.7: Distribution of\':H'iablc 4 (p2) ""·Number of generations Figure 4.8: Distribution of\'ariabk 5 (a3) \'S. I\ umber of generations 49 50 51 51 52 52 53 Figure 4.9: Distribution of Variable 6 (pd 's. ~umber or generations 53 f-igure 4.10 Distribution of Variable 7 (a4) 's. :-:umber of generati.Jns 54 l·igure 4.11 Distribution of Variable 8 (p4) ,·s. Number of generations 54 hgure 4.12: Distribution of Variable l) (aS) vs. Number of generations 55 Figure 4.13: Distribution or Variable 1 () (p5) VS. Number of generations 55 Fi!.!ure-LI·.l: Distributionof'Variable II (a(l) vs. Numbcrofgcncrations 56 ~. Figure 4.15: Distribution of Variable 12 (p(l) \'S. Number of generations 56 I·Jgure 4. 16: Distribution of Variable 13 (a7) 's. Number of generations 57 Fi1:-urc 4.1 7: Distribution of Variable 14 ( p7) vs. >-lumber of generations 57 1--igurc 4.18: Distribution of Variable 15 (aS)' s. Number of generations 58 Figure 4.19: Distribution of Variable 16 (pS) 's. !'\umber of generations .. 58 Figure 4 .20: Distribution of Variable 17 (a9) ,.s. ;"\umber of generations 59 Figure 4.21: Distribution of Variable 18 (p<)) ,.s. Number of generations 59 t Figure 5.1: Details of the factors considered in the modeV (l} Figure 5.2: Energy demand vs. Year (actual data and the model forecasted data) 65 Figure 5.3: Electrical Energy Demand forecast done with the model- Short . Term . , . 66 Figure 5.4: Load curve or Sri Lanka on I st .lun~ 2005 67 -- , ... ..,. .,. / Page 4 or 88 LIST OFT ABLES Table 1.1: Roulclte wheel selection procedure Table 2.1: CTB forecasts in year 2000- ~hort I cm1 Table 2.2: CEB forecasts in year ::'000- Long 'I cm1 f"able J.2: Fitness \'alues obtained with drfkrent parameter scuings 0 0 (based on data ti·01n year I 984 till yc:.~r 2000) f'ttblc 501: Comparison of the GA model lorcc.lsl and the Time trend forecast (done hy the CTB) for these two years. f'able 5.2: The possible Peak l:lcctricity Demand considering a 55% I ,oad Factor --/ " 14 20 27 44 j 64 •(J() _..,..: ..... .., / .. I P:tgc 5 of XS CHAPTER 1 Introduction 1.1 Concept of GA a11d cvo!ution:u·) progrm .• ;ning Our lives are essentially dominated by genes. ·1 hey govern our physical features, our bdw\ tor. our personalities, our health, and indeed our longevity [11. T.fie recent greater tmderstanding of genetics has pron~d to be a vital tool lor genetic engineering .1ppltcation'i in many discip li nes, in add ition to medicine and agriculture. It is wel l known that genes can be manipulated, controlled and even turned on and oiT in order to achieve dcstrablc amino acid sequences or a polypeptide chain. This significance discovery has led to the usc or genetic algorithms (CiA) for computational engineering ll ]. Genetic algorithms have been developed by John Holland, his colleagues, and his sludents at the University of Michigan. ·1 he goal of their research has been two fold: I. To abstract and rigorously C\plain the adaptiYe process of natural sysi.:.ns 2. To desi5rn artificial systems software that rct:11ns the important mechanisms of natural systems "~ .... !his approach has led to important discoveries in both natur:.d and at1ificial systems SCience l2J. l3 ). . ... (;' GJ\ presumes that the potential solution ol' any p~~oblcm is an inuividual and caJJa..b.e represented by a set or parameters. These parameters arc regarded as the l?,ciies of a chromosome and can be structun:J by a stnng or \alucs in hi nary form. A positive Yalu{ generally known as a fitness 'aluc. is used to reflect the degree of "goodness" of the problcmtflat woulu be highly related'' ith its objectJ\e \'alue. Page 6 of 88 (1/\s ~re search algorithms based on the meclwnics of' natural selection and natural genetic<>. They combine survival of the fittest among string structures with a structured ~c:t randomi1cd information cxchange to form a search algorithm with some of the intlnVsl>llll'S) ~ckttion ! Jo'itnc\s \1.11 111g l'nnl ( l' ·•rcnts) (;cnttir 1 Fitnc\\ Operation Suh·p11pulattnn tulf\ pr m cl ,.~ .,; f-igure 1.1: A Genetic /\lgorithm cycle Page 7 of 88 1.1.1 \ Vhat arc Gcnes'l In 1859 Charles D:.uwin (ISm-S?) publisheu an extremely controvc1:sial book "hose full title is "On the origin c~( spccws hy 111cans of natural selection. or the prcscrmtion offa\'Orcd mccs in rhc struggle for life", which is now popularly known as The origin of spcctes. I lc suggested that a species is continually developing, his controversial thesis implying that man himself came from ape-like stock. During his explorations. Darwin was impressed by the \'ariations between species. I Ie noticed that in almost all organisms there is a huge potential for the production of ortspring as. for example. eggs and spores. bu~ that only a small .1 percentage survi\ e to adulthood. lie also observed that within a population there is a great deal of variation. This led him to deduce that those variants which survived the struggle to adulthood were, presumnbly, the ones most fit to do so. Supposing that individual variation could he inherited by offspring, Darwin saw evolution as the natural selection of inheritable variations. Around the same time, Gregor Mendel ( 1822-84) in\'estigated the inheritance or char:.~cteristics, or trnits, in his experiments with pea plants. By e.xamining h~brids from different strains of plant he ohta111ed some notion of the interactions or characters. for example. when crossing tall plants with short ones, all the resulting h) hrius were tall regardless or \\'hich plant donated the pollen. \1cndel declared that the characters or genes as they later came to he kr1own, for the tall pi:Jnt was ,., dominant and that the gene for shortness was recessive. Although Mendel 's c'\pcrimcnts laid the foundations for the study of genetics, it was not until 30 years af'ter his death that Waller Sutton (1~77- 191<>) discovered that genes were _l?,.ilft or chromosomes in the nucleus. However, Darwin's theory emphasi;cd the role of continuous \'aria.tion within "' "' species. In contrast. disti net d i fferenccs bet ween species are not uncommol1 in nature, i.e. discontinuous \anat ion. llugo de Varis (1848-1935) obser,cd that in a population of cultivated pla1~:s. striktngly different \·ariants \\·ould occasionally appear. To explain this discontinuou-, \anation. de \'aris de\·eloped :.1 theory of Page 8 of b8 mutation. Superficially, the new science of g<.:nctics scemeu to support the mutation theory of evolution against orthouox Darwinism. With greater understanding of the structure of genes, genetics came to n.:ali;c Ito'' subtle the ertcct of mutation could he. If a characteristic is ddcrmined by a stnglc gene, mutation may have a dramatic effect; but ira battery of genes eombtncs to control that characteristic, mutation is one of them may only have a negligible ei'!Cct. It is clear. therefore. that there is not a sharp distinction between mutation and Darwmtan theory of e\ elution as they O\'erlap. The principle of selection uoes. hO\\e,·cr, remain sound. The fundamental unit of information in li,·ing system is the gcne,Jin general, a gene is defined as a portion of a chromosome that determines or aiTccts a single charuc ter or pheno type (visible property), I(H· example, eye colour. ll comprises a segment of deoxyribonucleic acid ( D~/\), commonly packaged into structures called chromosomes. This genetic information is capable of producing a functional btological product that is most often a protein. · 1.1.2 EYolutionary Computation Evolutionary Algorithms can be dt\ tdcd tnto three main areas of research [4]: Genetic Algorithms (GA) (from '' htch both Genetic Programming (which some researchers argue is a fourth main area) and Lcaming Cla~si~ Systems arc basco), ~ . Evolution Strategies (ES) and [, olutionary Progranuning. Genetic Programming began as a general model lor adaptive process but has become etTcctiv~ at optimitation while Evolution Strategies was designed from the beginning fox .•. • . . ' \'ariablc optimization. ~·- rhe origins of Evolution Computing can be tr:.tccd to early work by Co~nputcr Scicmists in the I 950s and I 960s \\'ith the iuca that evolutionary processes could be applied to engineering problems of optunit.ation. Tbis leu to three major P:.~ge 9 of 88 ~ independent implementations of Evolutionary Computing of which two arc Evolution Strategies and Genetic Algorithms. Genetic Algorithms were initially de\ eloped by Bremermann in I to the evolution of populations of individuals that an.: better su itcd to their el1\-ironlllclll than the individuals that they were created from. lndi' iduals or current appro\.imation-; arc encoded as strings. chromo:-omcs, composed O\er some alphabct(s). so that the genot) pes (chromosome 'alue~) arc uniquely mapped onto the dcti'>ion 'ariable (phenol) pic) domain. The most common!; used representation in Gas is the binary alphabet [0, I) although other representations can be used, e.g. tcrnar;. integer, real-valued etc. r:!r ei\ample, a problem with two variables, x1 and r_, , ma) be rnapp~.!d onto the chromosome structur\.! in the following way: 100 1 0110110 1 00101110110 1 X X' When?. x1 is encoded \\·ith ten bits and x~ \\ith 15 bits. possibly reflecting the level of accurac) or range of the indi\ idual decision 'ariablcs. E\.amining the chromosome string in isolation yield.., no information about the problem. \\hich \VC arc trying to sohc. It is unl) \\ith the decoding of the ,/ chromosome into its phcnot)pit: values that any meaning rmance, or fitness, of indi-vidual members of a population. This is done through an objective function that characteri/eS an indi\'idual's performance in the probkm domain. In the natural \\Orld. this would be an indi' idual's ability to sur\ivc in it'> present em ironment. Thus. the objccthc Page 13.of88 function establishes the basis lor selection or pairs or individuals, which will be mated together during reproduction. During the reproduction phase, each individual is assigned a fitness value dem ed from its raw perlormam:e measure given by the objectiYe function. This value is used in the selection proces-; to btas it to\\ ards fitter indi\ iduals. llighly lit indn iduals, relati,·e to the whole population. haYe a high probability of being selected for mating \\'hcreJs less fit indi\ iduals ha\ e a correspondingly low prooability of being selected. l Once the individuals have been assigned a fitness value, they can be chosen from the population, with a probability according to their relative fitness, and recombined to produce the next generation. Genetic operators manipulate the characters (genes) of the chromosomes directly, using the assumption that certain indi' idual's gene coclcs, on average, produce litter individuals. :\ scheme called Roulette \\'heel selection is one of the most common techniques being used for such proportionate selection mechanism. To illustrate this further. the selection procedure is listed in table 1.1. Table 1.1 Roulette n heel select ion procedure --/ " - Sum the fitness of all the population members; named as total fitness (F.111111), ~ - Generate a random number (11) between 0 and total fitness F,11111 • •• , - Return the first population member whose fitness, added to the fitness or the preceding population members, is greater than or equal to 11. .-•• /., For example, in figure 1.5. the circurnft.:rcncc of the Roulette wheel is F, .. 111 for all fiYe chromosomes. Chromosome 4 i~ the fittest chromosome and occupies the largest interval, whereas chromosome I ts the kast lit that corresponds to a smaller Page 14 of8S interval within the Roulette wheel. To se lect a chromosome, a random number is generated in the interval 10. F"'ml and the imlividual \vhusc segment spans the random number is selected. rill' cycle of evolution is n:peatctl until a desired termination criterion 1s reached. I his criterion can also he set by the number of evolution c:clcs (computational runs). or the amount of variation of indi\ iduals bct\\een different generations. or a pre-defined 'alue of litnc'>s. 01 1:12 03 04 •s rigurc 1.5: Roulette "heel selection j In order to facilitate the ( i \ e\ olution cycle, t\\o fundamental operators: ( 'rossovcr and mutation arc required. although the :;elt:ction routine can be termed <~'- the other operator. 'I o further i llustr.nc the opcrational procetlure. a one-point crossm er mechanism is depicted on figure I .6. A cru:;:;o\ er point is randomly set. I hc portions of the t\YO chromosomes beyond this cut-otT p~ to the right arc to be " C\changcd to form the ol'l~pring. An operation rate (pJ '' ith a typical valuc bd\\CCil 0.6 and 1.0 is normally used as the probability of crossover. crossover poinl .;' --·- .., Parents Offspring Figure 1.6: J: ,amplc of onc point crossOYcr Page 15 of88 ,/' llm'vever, lor mutntion (figure 1.7), thi s applied to each offspring. individually alter the crossover exercise. It alters each bit randomly \\ith a sma ll probability (p 111 ) with at~ pica! va lue of less than 0.1. Original Chromosome "C\\ Chromosome r:igure 1.7: Process of mutation The chose of p111 and P.: as the contro l parameters can be a complex nonlinear optimi7ation problem to solve. Furthermore, their settings are critically dependent upon the nature of the objective function. I his se lection issue sti ll remains open to sugge"t ion although some guide! incs have been introduced. - For large population si7e (I 00) ( rossm cr rate: 0.6 ;'vlutation rate: 0.00 I - For small population siLe (30) Crossover rate: 0.9 Mutation rate: 0.0 I 1.2 Electrica l energy demand forecasting .. / " .; --·- , hnecasting demand is both a science and an ~rt. The need and relevance of/ forccastmg demand for an electric utili!) has become a much-d1:,cussed ISsue 111 the recent past. I his has led to the developm...:nt of various ne\\ tools and methods for forecasting in the last two decades. In the past, straight-line c:\trapolations of historical Page 16 of 88 Cl!Crgy consumption trends served well. However, with the onset of inllation and rapidly 1ising energy prices, emergence of alternative fuels and technologies (in energy supply Jnd end-usc), changes in lifestyles, institutional changes etc, it has become very unportant to usc modeling techniques \\'hich capture the effect of factors such as prices, mcomc. population, technology and other economic, demographic, policy and technological variables[(>). There is an urgent need for precision in the demand forecasts. In the past, the world O\'cr, an underestimate was usually allended by setting up turbine generator plants fired hy cheap oil or gas, since they could be sd up in a short period of timeA·ith relati\cly small itn-estment. On the other hand, overestimates were corrected by demand growth. lhc underlying notion here was that in the worst case, there would be an excess capacity, \\'hich would be absorbed soon. roday an underestimate coukl lead to under capacity, v.hieh would result in poor quality of service including locali;ed brownouts, or even blackouts. An overestimate could lead to the authori?ation of a plant that may not be needed for several years. Many util ities do not cam enough to be able to cover such a cost with out offsetting revenues. ~1orcon:r. in ,·iew of the ongoing n.:fi.>nn process. with associated unbundling of electricity supply sen ices, tariff rclonns and nsmg role of the pri,ate sector. a realistic assessment of demand assumes ever-greater importance. These arc required not merely I li.lr ensuring optimal phasing of investments, a long-term ~hsideration, but also rationali/ing pricing structures and designing demand side management programs, which arc lll the nature of short- or medium-term needs. p The construction period for power plants, which arc S-0t up to meet consumer demand, typically \aries between 5 to 7 years in the case ol· thennal and hydro plants amLJ to~ ;.:cars lor gas-based plants. As a result, util1tics must rorecast demand for the long'nm (I 0 / to 20 years), make plans to construct l~tcilities and begin Jc\elopment \\ell before the indices of forecast grO\\th rc\'crse or slowdown. ~mce electric utilities arc basically dcdtcateJ to the objcctiYe of sen ing. consumc1 demands, in general the consumer can Page 17 of88 place a reasonable demand on the system in terms or quality of power. With some built-in rcser\e capacity, the utilities may h:l\c to conligurc a system to respond to these to the extent possible. In the process of making predictions, forccoster bcors in mind the feedback effects of pncing und other policy changes. and thcrcrore, participates in the process of designing ways and means to meet consumer demands. 1.3 P1·oposed methodology of electrica l en eq~y demand forecasting l There is an array of methods that arc available today for forecasting demand: 1. Time trenu method 11 End-usc methou 111. Econometric approach An appropriate method is chosen based on the nature of the uata available and the dcsucd nature anu level of detail of the forecasts. The proposed methodology is based on Genct1c Algorithms. The all-possible factors that af'fect the electrical energy demand arc con!:iidcreJ in designing the forecasting 111o~lel. By giving a particular weight to each factor & subjecti~to natural evolution a " mme accmatc electrical energy demand-forecasting model could be obtained . . ..-' Page 18 of88 CHAPTER 2 Survey of Available Electrical Energy Demand Forecasting l\tl ethodologies 2.1 TimcTr·cnd l\lethod This method falls under the category of the non-causal models .~f the demand li.m.:casting that do not explain how the values of the variable being projected are determined l6]. Here the variahles to be predicted arc expressed as a runction of time, rather than by relating it to other economic, demograph ic, policy & technological \Jnabks. This function of time is obtained as the !'unction that explains the available data. and is observed to be most suitable for short term projections. I he time trend method has the advantage of its simplicity ant.! case of usc. HO\\ ever, the ma111 disad,·antage of this approach l1es 111 the fact that it ih,rnores possible interaction of the variable under study "ith other economic factors. For example, the role of incomes. pnccs. population growth anti urbani;ation. policy changes etc .. are all ignored by the methot.l. The underlying notion of trend analysis is that, time is the factor determining the Yalue of the 'ariablc under study, or in other ~rds, the pattcm of the rariablc in the past will continue into the future. ,., lienee, it docs not of!Cr any scope to intcrnal i;e the changes in luctors .s~1ch as the cflccts of government policy (pricing or others), demographic trends, percapita income l'tc . fhm cn~r this method is important as it provides a preliminary estimat~ •. or 111 fotccastcd value of the variable. It may \\'ell serve as a useful cross check in th6 case of ., short -term forecasts. 8 ;j J (J J. Page 19 of88 The table 2. 1 is a forecast of electricity demand lor system planning requirements [7]. I hts is a short term projection done based on the Time trend methodology. Figure 2.1: Electrical Energy uemand forecasted by The CEI3 in year 200 I: Short- term prOJCCltons (done by the system planning brunch or the CEB) J'ahle 2.1 CEB forecasts in year· 2000- Short Term -..c. ~ .._ - "0 c "' E 0 >. .... 0 .... .... 0 w 12 10 8 6 4 2 Year 2002 2003 2004 2005 Forecasted Electricit) Demand (T\\11) 7.381 8.10(> 8.889 9.748 --/ ... -I.J 9.748 0 . ~. 2000.5 2001 2001.5 2002 2002.5 200} 2003.5 2004 2004.5 Year ,._ ..,- ·""' /"' Ftgun.: 2.1: Flcctricity dcnwnu forcc.tstcd hy ·1 he CEB in )T. 200 I: Short-tcm1 projections Page 20 or S8 2.2 End-use method The end-usc approach attempts to capture the impact of energy usage patterns or various devices and systems. The end-usc models lor electricity demand focus on 1ts \'arious uses in the residential, commercial, agriculture and industrial sectors of the economy U1J. For example, in the residential sector electricity is used lor cooking, air conditioning. refrigeration, lighting, and agriculture for lift irrigation. The end usc method is based on the idea that energy is required for the service that it delivers and not as a final good. The following relation detincs the end usc methodology for a sector: j E - Sx~xPxH E - Fncrgy consumption of an appliance in kWh s ~ penetration level (n terms or number or such appliances per customer) I' lllllllber 0 r customers P - po\\er required by the appliance in k\V H = hours of appliance use. "I his, when summed over different end-uses in a sector, gives the aggregate energy denwnd. Ibis method takes into account improvements in cflicicncy or energy use, uti ll/ation rates etc .. in a sector as these arc captured in the pm\ cr required by an I • appliance. P. In the process the approach completely captures tl1( price. income and other economic and policy efl"ccts as well. .. The end usc method is most effective \\hen new technolot:,ries and fuels have to be int roduced and when there is lack or adeyuate time-s~rics data on trends in consumption and other variables. I TO\\ ever, the approach demands a high le\ el of detail on eli~ of tile end-uses. One criticism raised against the method is that 1t may lead to niechanical l<)rCGlSting of demands, with OUt adequate regard for behavioral responses Of COilSUJ11CI"S. 1\l so. it docs not give regard to the variations in the consumption pattern due to demographic, socio-economic, or cultunll factors. A feature ol"this method is that the data Page 2 I of88 1s colkctcd \\'ith a picture ol· the end result in mind. For example, a stud y of the agriculture sector may require look at the area under each of the major crops, cultivation practices and water requirement per unit area irrigated (including percentage rain-fall) for these crops, etc. However, if one were to look at the agricultural sector as a whole, the degree of tktail required might not be a::; intensive. Therefore, the degree of detail n:qu1rcd in the data depends on the dcsm::J nature of the forecasts. · 2.3 Econometric approach l This approach combines economic theory with statistical methods to produce a sy::>t<:m of equa tions fo r fo recasting energy Jemand (6]. Taking time-series (detJilcd data o\'cr the last, some 20 to 25 years) or cross-sectional/pooled data (detailed data pookd over different regions/states/individuals and time as well), causal relationships (functional forms \\here a cause-and-cfTcct relationship is establi shed between variables. Eg., rhang.cs in income cause a change in consumption anu/or vice versa)coulu be established between electricity demand and other economic variahles. The depen<..lant variable, in this case .. ucmand for elcctricit~. is e\prcssed as a function of various economic factors. These \ ariables could be population, income per capita, price of pov>er etc. Thus one would ha\·e: Where, DE= f(Y. Pi. POP) ED =electricity demand '{ :::- income per capita Pi - price of power POP population --/ "' . ' :.4.. , . ~· .. ' Se' ci·,d functional forms and combinations of these and other variables m'!'!y have 1;9' be 111ed till basic assumptions of the model arc met an<..! the relationship is found statistically signi fie ani. Page 22 of 88 For example. the demand for energy in sp<..:c ili c sec tors could be explained as a runction of the variables indicated in the right hand side of the following equations: Residential ED f(Y of per c;.:pita, POP, Pi) Industrial ED ~ g (Y of power 111tens1\C industries, index ofT, index ofGP) \\'here, T ~ tcchnolog) GP = government poliC) Inserting forecasts of the independent vari;:blcs into the equation would yield the projections of electricity demand. The sign and the cocrticicnts or each variable, thus cs t i rn ~rted, would indicate the direction and strength of each of tiJ right-hand-side \'ari~rhle in explaining the demand in a sc<:tor. l'hL: econometric rncthod requires a consistent set of information over a reasonably long duration. This requirement forms a pre-requisite for establishing both sho1t-term and long-term relationships between the variables involved. Thus, for instance, if one were in terested in knowing the price elasticity of demaml. it is hard to arrive at any meaningful cstunatcs, given the long period or administered tariffs and supply bottlenecks. I lowe\·er, the pnce eff"cct will ha' e an important role to play in the years to come. In such a c:.~se, one may hJve to broaden the set of explanatory variables apart from rcl) ing more 1 igo rous econometric techniques to get arot.1~J the problem. Another criticism of this mcthoJ is that during the process or forecasting it is incorrect to assume a particular grow th rate for the explan~tory variable. Further, the approa~ fails to incorporate or capture. in any way, the role of certain policy measures/economic shocks that might otherwise result in a change in the bd1avior of the v<.~riablc being explained. This would h:t\'e to be buill into the model, maybe in the form of structura l changes. ... ..... _ 2.4 ( ombining econometric and time ~cries models ..,- / .., It 1s common to use a combinat1on or econometric and time series model to achie\e ·greater prec1sron in the forecasts. lhis ha!:> the ad..,antage of establishing casual Page 23 of 88 rci;Jtionshirs as an econometric model along with the <..lepcndency relationsh ip [6]. Various func tional forms such as linear, yu;Jdratic, long-linear, translog, etc. arc used to capture the possible trends that m~ty be C\ idcnt in the data. The functional form of the model is derived at after a trial am.! enor process. A model is built using the available dat:::. truncating the last lew obscn at ions. The procedure for testing the model entails making prc<..lictions for the last fc\\' time pcnods for'' llich actual data arc a\ ailahlc and were truncated. The functional form \\here the forecasts ha\ e least <..le\ iations from the data availabic is chosen. l 2.5 Load f01·ecast for 2001 Generation Expansion planning Studies with Econometric Method The general multiple linear regression model used in econometric analysis is of the form : 1; = h, +h!X,, +h,X,, t t\ +h4 X 4, +e, Where. Y is the dependant \ ariablc; the X's are the independent variables. c, the error term and b, is the constant tcm1 or intercept of the equation. (X2, represents for example, the 11'1 observation on explanatory' ariablc X2) !8]. .,.1 In vic\\' of the above discussion, the following mo(cis \vcrc used to analyse the demand bcha,ior of different sectors under investigation. ,.,, 2.5.1 Domestic sector -...~ ... I)(!), - h1 + b~ Xo [)( '(t I), +h 1 !J(t - I), -t- e, ..,- \Vhcre, D(t), - Demand for electricity in domestic consumer categv. j XoD( '(t -I) -?\umber or domestic consumers in JXC\ ious year Page 24 of88 D(t -1), - Demand in domcst ic consumer category in previous year 2.5.2 Industr-ial and Commercial Sect<))· D(t) = h, + h~GDP + h,(J DP(t ·I), + h.1 IJ(t- I), + e, Where, I J(t) -Demand for electnc1ty in Ind. And Commercial consumer categories .I GDP - Gross Domestic Product (i !JP(t I), -Gross Domestic Product in previous year D(t I), - Demand in previous year 2.5.3 Other sector The other two consumer categories \\h1ch do not fall into any of the above two cJtcgories arc considered in the ·other sector' '' hich includes sales to religious purpose consumers and that consumed in ~trcct lighting. Because of the di\crsc na~ure or the consumers incluucu in tim; catcg01y, it is better to analy/c this category with out any links to other social or demographic \'aria~les. Hence, a time- trend ".r analysis \\aS used to predict the demand in thi<> sector. For the time trend analysis, the following linear regress ion equation ( I) )"as derived starting from the relationship: Where sAu,·s,=b,(J+gJt g - the a\'erage annuJI growth rate b, -constant 1:3 - constant ~ ..... ..,. ,/' Page 25 of 88 ln(SA !,F.S) 1 H 1 In( I l g) I -----------------( I ) 'I he table 2.2 gi\·es the forecasted electricity demaml for the usc of generation planning branch in the CEB in year 2\JOO, based on the Econometric approach. The 2'1'1 column shows the C"Xpcctcd clccuicit~ tk:nand (for consumption) as Low, ~-kdium ami lligh values till year 2020. !"he expected system losses are gi\'en in the ne"\t column, as a percentage. The requin;tl generation to meet the total tlcmand is presented next. Considering a Load F::~ctor of 54.2°o, the possible peak demand of the year is calculated and gi\'en (as Low, Medn11n anJ lligh values) irAhe fina l column . .. / ... ·"" ...... ..,- , ... Page 26 of 88 T ah k 2 .2 C E B f or ec as t\ in ~e ar 2 00 0- L on g T er m - - - - - Sy ste m Lo ad Y ea r D em an d (G W h) Lo ss es (% ) G en er ati on (G W h ~Fa ctor (%) f lo w Po ak (M W) - [H igh !M ed ium - . l J!: _o w M ed iu m Lo w M ed ium Hi h H1 gh - - j5 4. 2o % · 14 50 . ' - - - - 1 12 00 0 54 16 54 16 ~~~ 21 .4 0% 68 86 16 88 6 68 86 11 45 0 14 50 12 00 1 57 48 57 48 8 19 .3 0% 71 19 71 19 71 19 54 .2 0% 14 91 14 91 14 91 20 02 61 91 62 34 62 78 18 .50 % 75 94 17 64 7 77 01 54 .2 0% 15 91 16 02 16 13 12 00 3 66 42 67 59 68 75 17 .8 0% 80 80 82 22 83 63 15 4.2 0% 16 92 17 22 17 52 20 04 ,71 06 73 01 74 96 16 8 0% 85 41 87 75 90 09 54 .2 0% 11 78 9 18 38 18 87 20 05 17 58 5 78 65 81 45 16 .1 0% 90 42 193 76 97 10 54 .2 0% 18 94 19 64 20 34 '20 06 80 83 !8 45 0 88 16 15 .4 0% 95 49 !9 98 2 r0 41 5 54 2 0% 20 00 20 91 2• 81 I 20 07 . 185 81 90 36 94 92 14 .9 0% 10 08 0 10 61 5 11 15 1 54 .2 0% 21 11 22 23 23 36 20 08 ~90 83 96 32 10 18 1 14 .8 0% 10 66 0 11 30 4 11 94 8 54 2 0% 22 33 t23 68 25 03 20 09 95 91 10 24 0 10 88 9 14 .6 0% 11 23 2 11 99 3 12 75 3 54 .2 0% 23 53 . 25 12 26 71 20 10 10 10 9 10 86 5 11 62 2 14 .6 0% 11 83 7 12 72 2 13 60 8 ' 54 .2 0% 24 79 1 2 66 5 28 50 20 11 10 63 9 11 51 1 12 38 4 13 .8 0% 12 34 2 13 35 4 14 36 6 15 4.2 0% 25 85 27 97 30 09 12 01 2 11 19 5 11 21 94 11 31 93 13 .50 % 11 29 49 14 10 5 11 52 60 54 .2 0% 27 12 29 54 31 96 20 13 11 76 8 12 90 3 14 03 9 , , . . . 11 3. 30 % 13 57 3 14 88 2 16 19 3 54 .2 0% 28 43 31 17 33 92 20 14 11 23 57 13 64 1 14 92 5 1 1 3. 10 % 14 21 3 15 69 0 17 16 7 15 4.2 0% 29 77 32 86 35 96 20 15 12 96 4 14 40 9 15 85 5 12 .8 0% 14 87 0 16 52 7 18 18 6 54 .2 0% 31 15 34 62 38 09 20 16 r3 59 1 15 21 1 16 83 1 12 .6 0% 15 54 7 17 40 0 19 25 3 154 .~ % 32 56 36 45 40 33 20 17 14 23 7 16 04 7 17 85 7 12 .3 0% 16 24 2 18 30 7 20 37 2 15 4.2 0% 34 02 38 35 42 67 12 01 8 14 90 4~ \ 16 92 1 18 93 7 12 .1 0% 16 95 9 19 25 4 21 54 8 54 .2 0% 35 52 40 33 45 13 t 20 19 15 59 4 17 83 5 20 07 5 11 .9 0% 17 69 8 12 02 41 22 ~5 4. 20 % 37 07 42 40 47 72 l~Q 20 ,1 63 07 21 27 5 I 18 45 9 21 27 1 ,2_ 10 83 54 .2 0% 38 66 ~ 5 50 44 · 11 .7 0% ' Pa ::;e 2 7 o f 8 8 -- l·igure 2.2 shows the e lectrical energy demand of Sri Lanka from year 1985 till year 2000 191. 1101. Electrical Energy Demand of Sri Lanka vs. Year 6 4 2 ? ~ ..._ - -o c ro E Q) Cl >. O'l ..... Q) c w //··- 1984 ... ,....,...., • . / ' / . / I ~ 1990 Year -.- , ; , , , ; ,,/ ,' ~ , ,/\/ . j _.:I . , . ~ ~ ".r 2000 I it:ure 2.2: The elect rical energy demand or Sri Lanka from year 1985 till year 200~ · In the above figure (Figure 2.2) during the year~ 1996, 1997. l l)98 and 2000 a s1gn1ficanl demand reduction could be observed. this may be because or the po~~er-cuts unpn-;ed hy the CEB during those years. In 1)0's ~ri Lanka \\as a countr) main!) depended on hydro-pm,cr. As the only dcctric utility available in the country this should nc,cr happen. The demand forecasting, Page 28 of88 ll , - - ... c.j ~ planning, implementation, as well as decision muking, at correct time in an efficient way IS a must. Otherwise the outcome would be ' cry pathetic. Long ~nd heavy droughts e\pcricnccd in above mentioned years limited the hydropower generation. The capacit) or thcrn1:1l power plants connected to the system could not meet the dcnund. So the result ''as power cuts on public consumption. In certain years this led to blackouts even As the authority that caters electricity to the country 1t is their duty (as well as a responsibility) to maintain the quality and reliability of power. That should nc,er be limited to a couple of words. As consumers, people pay for clcctncity not only for their consumption; but abo expecting a high qGality of service liom the ut ility. If the utility was up to its plans this would ne,er hapJ'>en. It is a we ll known lill't that incnicicncy and poor dec ision mak ing of top management of the CEB arc the major reasons for th is nationa l crisis lll J, l l2J, [ 13]. As a country, lack (or may be its scarcity) of' ava ilabi li ty of energy/electricity is a maJ or factor for po\ erty. So the authorities and the personals of decision making in this sector an.' bound to act efficient! y. ciYcct i' cl y, and 111tcl I igent ly in the march towards its future goals. lienee, there is an ur.;cnt need for precision in the demand forecasts. Today an under cstinlJte could lead to under capacity, which \\Otild result in poor quality of service ,,,1 mcluding localized brownouts, or e\cn blackouts. An ovcrcs(r;u1te could lead to the authori;ation or a plant that may not be needed for several years. ,.. Figure 2 . .'\ shows the elec tri city dc nwnd rorccaslcd by the (jencration planning branch of the CF B in year 200 I [9], [I 0]. #f . .. ..;- / ... Page 29 of 88 Electrical Energy Demand- CEB forecast 2001 .......... ..c ~ 200001 __. "'0 c ro E (1) 0 >. ..... () I... ..... () Q) w 10000 2000 p· Medium Low , __,/ _, // 2010 2020 Year .. / Figun.: 2.3: Electrical Energy dcmantl rorccastcd by The CEB in ~ear 200 I: Long term .~ -~-- ... y / .., Page 30 of88 CHAPTER 3 Proposed ''lethodology of Forecasting Electrical Energy Demand based on Genetic Algorithms 3.1 Pt·oposcd :\lethodology f ... l'his is a llC\\' methodology with which the Electricity Demand or a particular system could he f'orecastcd with a higher accuracy. In thi s method all possible !actors that arJcct tiH: ckct rica! energy demand of the system such as time, population, population gro'A th rate. (iDP per capita. average US S \~.due (in Sri Lankan rupees), number of domestic ronsu:ners. average electricity price. rain fall, other institutional factors (policy & technological variable) etc. are considered. Cons1dering all possible factors. dem c and repn.:scnt genetically to a model for l:lcctrical Energy Demand forecasting. ·1 he Ftgurc 3.1 demonstrates the representation of <.Jcncs in the Genetic Algorithm. ../ " .~ Page 31 of 88 (;l)pxl,.(i= 1,2, .... ,N) PRICE OF ELF:CTRCITY x 1, .(i == 1.2 .... ;\') ~ Elements of the Fi fncss function F(x,),(i 1,2, ... ,N) OTHF:R PARA.\lETERS X 11 • (I = 1.2 .... • \/) (i::.l.2 .... i\') .J M- number of factors under con~ideration N number or parameters Figure :l.l: Representation of genes B) gtving a certain weight to each factor :md subjecting to n~tural evolution. a more accurate forecasting model could he obtained. as bricll:d in hgu;.{3_2. .~ {\1( ll{l i\CTt IR/\11' + ........ ..,. ~101>1.:1 Figure 3.2: \\ ay to obtain the forecasting model . / ., P o f~or:n ~snt• ~ f .wn \\'here . ./~:c >FWC I'\!W = f(x) .. ,. " To avoid negative values, 1 ( . )' .fuuwJ?- = f 1c .,, '.II. - ./1·11/11 ( ·. I.' II 11 - ---- ------ ------ (2) · -"' --·- \\'hen .fi uuoR' is plot against the number of generations, a graph as shown in Figure 3.3 / \\ould he recci\ cd. Page :n of88 f I i(Ji( Ji• -....... ___ ___ ~o. ui' !..!Ctll:rat i nn~ j Figure 3.3: Distribution or li Ufl()f/ vs. :"-lumber or generations In thi" analysis, actual data for each factor have been considcn.:d from year 1984 to year 2000. (I) & (2 ) => /,.IW•/1- =- (f, , II' 11 - .~ n/1/l'l' ) " , ~ + (/, 77 t1 - /~0/IH 4St ·, )t <>s< ~ + """""" + (/~1 /l II - .l~-(}/1/.( ,l.\11 I t :'\ tnc factors ha\e been consitkrcJ in this model. lienee the Genes of the Genetic Algorithm ''ould be these factors. 0 w 0:: C/'lt.L x,..., 2~ 07. <0 .;....{._., ..-:, "' :\vc. \nnu.tl Rainfal l in cntchmc.:nt areas (iiW l'npulatio1t l'npulatinn gro\\'th rat>: •\\Cra)!C l ~ ", \.lllll~ I )ttmcsltc C.:illl~tunct · a~:counts \\ C unit pri.;c of UOilH: ... tic COI1'>11111ption \\c. unit pm:..: or lndu-;, & Com. C(lll~lllllp. \\c. unit prk c ,,l l'lhcr constllllption ,\c..:.., . Figure 3.4: L.tctors consider~. ... : in the forecasting model .~ .... ..,. Page 34 of X8 / ., 3.5 .Justification of selecting the considered factors \\'hl:n rc ICrrcd to the Sri Lankan conte:-.t, there arc several factors that affect the clcctricit;, demand significantly. I he l'igurc 3.5 and l"igun: ).6 sho" the consumption of cb:tricity by main consumer categoric-. of ~ri Lanl-.a and that as a percentage of the total consumptil'n of the )Car, referring to the ~ca1" 1990, 199), 2002 and 200.3 data. -~ ~ C> 35001 - -g 3000 ro 25001 ~ 20001 z-1500 1 :g 1oool ...... (.) 500 Q) W 0 Jl l.;;,;J I I aill Domestic Other lnd.&Com. Consumer Category j oYear 1990 · ll3Year 1995 o Year 2002 . o Year 2003 Figure 1.5: Consumption or electric it~ by main consumer categories of Sri ' ~mka 60 ~ so 1 ro (/) 40 1 ro "0 30 t: ro E 20 Q) 0 10 1 I 0 ' I KUI I I [}m . . Domestic Other lnd.&Com. Consumer Category .. / ... oYear 1990 •Year1995 oYear 2002 oYear 2003 #f . ... " .... / .., I igun: 3.6: Consumption of clectricit) b) main consumer categories of Sri Lanka as a percentage ofthc total con->umcr~ of that )Car Page 35 ofg8 l'he above information clearly shows that the demand lor electricity by the Industrial and wmmercial sector consumers. increase~ annually compared to that of other sectors. Our electricity <;uppl) is a system in tran<;iti\)n. The fuel mix in the generating system is rapid!) changing from a predominant!) hydroelectric o;y~tem to a predominantly thermal system. and Sri Lanka is unable to manage this transition Ill ].1121.1131. rhc :-.tructure of clcctricit) consumption i~ also changing from a consumption pattern do minated by the manufacturing industr). to one in "hich households and commercial builtllngs u~e more than 48% ofelectricit) sold (l ·igurc 3.7- Year 2003 data). j Industrial consumers 34 78% Commercial consumers 16.78% ~---.---- Domestic consumers 32.14% Other consumers 16.30% J'igure 3.7: Consumption of electricity by CommerciaL Domestic. Industria! an.J Other consumers in ~ri Lanka in ye,lr 2001. I ',ri I anka electricity generation sct:tor has been domina~ by hydroelectricity for man; )Cars. With the saturation of economically exploitable hydropower capacity and in the absence of any other reliable incligcnou" primary energy sources that can be useJ for ' large scale electricity generation. Sri Lanka will have to rely heavily on thermal generation based on imported ro~sil rucls 11-t 1. ..; /~ 3.5.1 Rainfall data in catchments areas \\hen rainfall is considered as a l ~1ctor \\hich affects the clcctricit) demand .of \ri Lanka. one can look at it in ~ever~! point-of-\ ic\\S such as. Page 36 of 88 • the direct impact on the consumption of electricity due to the rai nfall i tsel r, • the cost of energy due to the variations in the usage of fuel in electricity generation. Out of the aboYe two, the second one is more out-smart. The more the rainfall \\'C get higher the hydropower generation. lienee the cost of generation of an energy unit becomes less (e.g. Because of' the heavy droughts occurred in year 1996 and year 2000, the Sri Lankan power utihty was unable to meet the demand. with the a\ailable hydro and thermal po\' .. er plants. This caused localizecV'brown-outs and C\Cn black-outs.). Since there arc limitations in collecting rainfall data all over the country, the average rainfall ligures in catchments areas that have a considerable impact on the hydropower generation have been considered. 2600 ...-. E E ..._... en ~ 2400r c Q) ~ ' 1\ I \ 1 \ "r E .s:: (..) .._. ro (..) ~ 2200 ro ..... I I I '\ I .. II I ' .6- c ro 0:: 2000 -~ ..... ~-~__.-~~- ,___.__ y ~ / 1990 1995 2000 1985 Year Figure 3.8: Rain!'all in catchment areas Page 37 of88 The past data shows that the rainfall in Sri Lanka gradually reduces [ 15), (Figure 3.14: Rainfall in catchment areas). So even though the installed hydro capacity is much higher, it may not cater the system fully due to lack of rainfall. li enee special attention \\"aS paid to :ma\~;c the ciTcct of rainfall un supply!dcmand 3.5.2 Domestic consumer accounts i\ \though there arc 'cry large industrial consumers connected to the system, a considerable portion of the electricity demand in St i Lanka is due to the domestic consumers (More or less the night peak is due to the clectricit/consumption in household.). So they could he considcn:d as an clement, which _decides the energy dcmaml of the country. The Major component of the Number of Electricity Consumers in Sri Lanka !'ails in to the category of Domestic Consumers. In Sri Lankan power system, the peak demand occurs around 8.30 p.m. (Typical example- Figure 5.4: Load curve of Sri Lanka l 't June 2005). In fact the domestic consumption is the reason for the night peak. Hence it performs a major role in deciding the energy demand of Sri Lank.a. --~ " 3.5.3 Average US S value With the onset of intlation, rapidly rising energy prices, development project done under foreign aids (loans). etc. the value of the rupee reduces day b~ clay compared to that of the US$. This US S value is more stable in the market. Sri Lanka docs not have its own oil wells to pro\'idc fuel fo!'•-clcctricity ~ generation. When dealing '" ith foreign market in purchasing fuel. it is m6re com·cnient (e\cn to do a comparison), ""hen it is giYen in terms of US S rather than Ill rupees. Page 38 of 88 The price of the f'ucl imported affects the electricity generation and hence the tariff system. This influences the energy usage and the energy usage patterns of the consumer. Therefore the a\eragc LS S \a\ue in rupees could be considered as a maJor factor that affects the clcctrictty demand of Sri Lanka. 3.5.-l Population, Population gnm th rate \Vith the increase of' the populauon and population growth rate, the electricity j demand increases. This is due to reasons such as • the energy demand directly exerts on the system by the consumers during there house hold activities • introduction or new industries and their developments to cater the population l knee the population and population growth rate have been considered as ractors '' hich af'rect the electricity demand of Sri Lanka. The Figure 3.9 shows the Lstimated miJ year population against year for the past 22 years, [16). Estimated mid year popula tion vs. Year .. / " 25.000 c 0 ·~ ~ 20,000 :l a. 0 a. .... 15,000 co Cl> >- '0 ·E 1o.ooo 2 co E 5,000 ........ '.;:l (/) I w ..; /~ 0 1980 1985 1990 1995 2000 2005 Year Figure 3.9: Estimated mid year popul:ltion 's. Year l'.1gc 39 of 88 3.5.5 GDP Gross Domestic Product (GOP) is the unduplicated value of all goods am\ sen ices produced in a year within Sri Lanka's boarders at market prices. It is the standard or the 0\'Cra\1 si7C or the economy. The market value of final goods and services produced oYer time including the tncomc of foreign c01vorations ami forctgn residents \\orking in Sri Lanka, but excluding the income of Sri Lankan n:stdenls and corporations o\'crseas. Increasing energy consumption ' the development or the clcctftcity sector is strongly coupled with the economic growth of Sri Lanka (141. The Figure 3.10 and hgurc 3.11 show "the variation of Economic Growth and Demand for Energy in Sri Lanka" and "the variation of the GOP in US $ against year" during the past 22 years. '•"!. l\" f . .. f\ . ., ;; f'\ (t'•n I '·z -~ :-:! j ., ., CY. .. r. '" ; J •t u ~ _, )" : .. . lr'u t ·--··- 20', I I , . ·' I 1t1 ~~~ .. . .. ,· , ' ~· .. 1 :• , . I\.. ·r . . <;· ~· • 'J .... ..;, 1, ... • '\ •· • 81o >, \. I I ._ / ., 1\• \1'1· " , I . 1 '\} , ..} I ·w .., \ . I I t ~:.. 10 ~ \' 4 °1 I I .r :'. r, '/ ' --~ I -4"' ' .-•"'\: /0 r ~ ..J ... . ,.. I ,~ 1 -f .1 0(~~ • .. ' :r , . r '. (~ Ye:\r lll ll 1 • • • I Itt II ii.'lt' ... I ~ ~·- ..; \ ..... ~ .r Figure :\.10: The variation ol' fconomic GrO\\ th and Demand for Energy m Sri Lanka. '~ Page 40 or 88 GOP in US$ vs. Year 1000 900 -~ 800 700 _J/ tl) (/) 600 ::J :::: 500 0.. 0 400 (.!) 300 j 200 100 0 1980 1985 1990 Year 1995 2000 2005 Figure 3.11: The variation or the GOP in US$ against year 3.5.6 Price of Electricit~· (Dome~tic) I he increasing electricity consumption ts abo strongl) coupled with. the price of electricity [ 14]. lt is a ''ell known fact that people purchase commodities when C\ er that commodity is more 'a1uable to them than its price. l~oday to get our work done .. ., energy is essential. As the most com cnicnt way or receiving energy is electricity, it has a very high Jcmand. The inability to meet the demand is the reason [or the power crisis we are experiencing today in Sri Lanka. ,.6t '• The unit price of electricity is a main ··ractor that decides the demand of • electricity. When consider the pO\\CI' system or Sri Lanka basically the'·elcctricity /~ consumers could be di' idcd in to three major categories as • Domestic consunH.:rs • Industrial and cotmm.:rctal consumers • Religious and other pmposc consumer. Page 41 o;' S8 1\s discussed before, the demnnd by the Domestic Consumers is a main reason lor the night peak. /\s a whole these consumers do not consume bulk power. Their COilSIItnption is just tO Satisfy their li:ly to day requirements. lienee the number Of dome~ttc consumers as well as the a\ eragc price of such a domestic consumer could be l.'(ltlstdercd as the factors that a!Tcct the elcctrictty demand of Sri Lanka. \\1Jen refer to the Industrial ami commerctal sector consumers, although they arc much less in quantity, the demand tilt:)' e\.ert on the system is very high. Since the number of units they consume' aries'' ith in a l:.H6e range, number of industrial ami C ommcrcial consumers could not be counted as a factor that affc" on an) nc" i nd i' i d ua \; arc create~ :'\umber or variables Precision of binary rcprescntntion tJppl.:r and lower limits (range) or(/ and,;[_ (Field descriptor) J ,\Iter 20. 000 no. of generations. fruuo/ ·~ 0.6775375 - 500 20.000 = 0.9 =- \8 --~20 - bet \Vecn -2 ~nd + 2 -~-- ... J _,., \\'ithin the first 5000 generations the ,·ariation or the best fitness was significant, (I· igurc: 4. I). Page47 ot'88 Distribution of Best Fitness in first 1000 Generations 4 3 51 30 .J en 25 en C1) c: :-= 20 u.. ~ en C1) 1.5 co 1 0 \ ~ 0.5 0 0 1000 2000 3000 4000 5000 6000 Number of Generations F1gure 4.1: Sample f[:RJw/ \ s. No. of generations-distribution with I 000 generations --/ " .t.2 Results used in the forecasting Model .. ; '. The graph in Figure 4.2 gives the Distribution-of Best fitness value vs. number or ·~ ).!Cilcrations. /\ rter 20. 000 generations 0.6775~75 was received a<; the best fit ~~luc . ..; Page 48 or 88 (/) (/) (.) c .... u. .... (/) (]) Ol " 35 3 2 51 2 1 5 05 0 Distribution Best Fitness vs. Number of Generations ~ . .r 0 5000 10000 15000 20000 25000 Number of Generations Figure 4.2: Di~tribution of bc-.t litnc-.~ '-.. no. or generations Cnnsidcring the Best fit value ~ 0.6775175. ..~ I. ......... , 6 "' a, and p, in (I). \\'here \\t:rc set to the following values that were recci'ved at the best fit condition after :20.000 generations in order to obtain the Model I~H· Forecasting Electrical Energy . .. I kmand or Sri l.anka. ..,. ~ / Page 49 of 88 (II \ r+ -1. 18024 <12 \ 0.742877 a\ \ 0.128874 (l,t \ ... -0.92-\89 a.; 1.0Y3729 ao 1.98654 a; ... 1.557174 as -0.4403~ a') • -1.(> 1072 PI ..... -1.31335 P2 ... 0.048903 I l P> ... -1.57()1 P·t I ... -1.51104 rs -, -.37124 P<> I -1.77698 P7 I 0.452536 px 7 • O.OlW473 P<~ -1.4893 Figure 4.3: Parameters obtained at the bc~t-lit 'alue .t.J Graphical Representation of e\ olution of t he parameters I I hi.! follt)\\ing graphs (from Figure 4A up to Figure 4.21) ~(c the evolution of these parameters. and some of them clearly show their convergence (either a1 or p1 values) to a <.;tC ·1.5 ·2 Figurt.: 4.4: Distribution or Variable I (a 1) vs. Number of Qcnerations 5000 10000 15000 20000 --~ " JIL f I L J~ _f \_jlf- Number of Generations 25000 :.._ ,;· , ·""""- ·~ , ~ / Figure -L5: Di:::.tribution of Variable 2 (p,) vs. ;-..lumber or generations Page 51 of88 2 1') N :.___r - I. (II U'-~ ~_.~ ..... 0 0.5 Cll ::::J (II > 0 5000 10000 15000 20000 25000 05 1 j 1 <; Number of Generations Figure 4.6: Distribution of Variable 3 (a~) v:,. Number of generations 05 OJL ~ r-v .. a. ..... 0 Cll 0.5 ::::J (II > 1 5 2 5000 10000 15000 20000 25000 ..~ .. ..,. Number of Generations figure .f.7: Distribution of Variable .f (p1) \S. '\umber of generations Page 52 of" 88 1 5 05 fl~ '1 - ~" 0 j II '--' ' ... 5000 10000 nl \....-- r _.," '-.------ 15000 20000 25000 -0 ~ -0.5 nl > -1 .J -1 5 2 Number of Generations Figure 4.8: Distribution of Variable 5 (a3) vs. Number of generations 2 I 1 s l n '1 .-r- - ·1 fl I ~ 05 1 f~n~l :I I I I ~ 0 til II ~ I 5000 I ll 10000 I 15000 I I 20000 25000 Cll :::s ~ 05 1111 11,1 I ll I I ""' L l f 'l -1 5 I II I I H [_ ........ ..,- ~ -2 / 25 Number of generations Figure -L9: Distribution of Variable 6 (p 3) 's. Number of generations Page 53 of 88 0 0 ,, 5000 10000 15000 20000 25000 ... -';jll-LL LJL_.- l ,-- .__ r~- i -1.5 -2 Number of Generations Figure 4.10 D istribution of Variable 7 (a.,) vs. Number of generations 1 5 0.5 a\ 5000 10000 -~ 15000 " 20000 25000 ... a. .,__ 0 -0 5 Q) :J -1 111 C1l ,, •' > ~~I I rd -- t_,__.J~ -1 511 lr-t\ L yl -2 /~ -2 5 Number of Generations Figure 4.11 Distribution of V,triablc S (p.t) vs. ).umber of generations Page 54 ofS8 2 1 5 ,\Ill 'I~ r ., 05 111 ' J l ' L L "' - _f l l 0 c:> :J 0 l.i nl > 'I 5000 10000 15000 20000 25000 05 I i ·1.5 ·2 Number of Generations figure 4.12: Distribution of Variable<) (as) vs. Number of generations 1 5 05 .. ~ 0. 5000 10000 .: 15000 20000 25000 ., I I a. I ~ -0 ·0 5 (),) t :J "' ,,rL 1 > ., 5 ' I - l I ·"' ,r -2 /~ -2 5 Number of Generations Figure 4.13: Distribution of Variable I 0 (p5) vs. ~umber of generation~ Page 55 of 88 2 1 51 05 "' ol C'O 5000 10000 15000 20000 25000 - I 0 ~ .Q 5 ::::s C'O > 1 Ill I -1.5 11 1 \ ll -2 l I --· f t._ u \___f- \ _j l 25 Number of Generations Figure 4.14: Distribution of Variable 11 (a6 ) vs. Number of generations 05 .. ~ .,. 0 II -os•l 5000 10000 15000 20000 25000 .., a. -0 Q) ::::s -1 C'O > 1 5 1-)) rr-,- L J J~_.f -2 -2 5 Number of Generations Figure 4. I 5: Distribution of Variable 12 (p~,) YS. '\umber of generations Page So of 88 .... 10 -0 Q) ::J 10 > 0: -0 Q) ::J 10 > 25 2 , ______ _ 1\l: 1rr L __ n-- 1 5 OS 0 5000 10000 15000 20000 25000 -1 I -1 5 ·2 ·2 5 Number of Generations Figure 4 .16: Distribution of Variable 13 (a7) vs. Number of generations 1.5 0.5~ .Jl~ -~,___ 0~ 5000 10000 15000 OS · 1 ___J" "!,.- " 20000 25000 . ..... · ·~ ,- ~ 1 5 / -2 Number of Generations Figure 4 .17: Distribution of Variable 14 (p 1 ) vs. Number or generations Page 57 or 88 25 2 1 5 ., ('3 -0 0 ::J 05 111 > I 0 m_ soo:_rl 10000 15000 20000 l 25000 -0 5 -1 Number of Generations Figure 4.18: Distribution of Variable IS (as) vs. Number of generations 05 I Old uu r-~I' I 5000 I 10000 15000 20000 25000 I 0 5111 - _r-~ I ,_~ Ill ., Cl. -0 -1 ::J .,., 111 > 15 I ·2 ~ ........ ,.,- ~ / 25 Number of Generations Figure 4.19: Distribution of Variable 16 (ps) vs. -:\umber of generations Page 58 of 88 Figure 4.20: Distribution of Variable 17 (a•J) vs. Number of generations 25 2 1 sl OS --/' " "' a. .... 0 0 -051' 5000 QJ :l ro > _, ·1'1 t-n 10000 15000 20000 25000 . i".~ Jll -:. ,_., 1 5 ~·- ..,- ~ ·2 / 25 Number of Generations Figure -L21: Distributton of Variable ll) (pi~) vs. Number of generations Page 59 of XX Accordi ng to the Figures -l.7 and -l.l7, it could he considered that the parameters p 1 & p-: have come to a steady state condition after 20, 000 number of generations. But "hen refer to figures such as Figure 4.9, pa ra meter p3 has not been converged to a 'teady state condition ami it doc!'> not show even a n ind ication of its con' crgence. The evolution of the parameters p 1 (Figure -l.5), a3 (Figure 4.8) and P<· (Figure -l.l5) show they \\Ould cotncrgc to some <,tcady ~tate level "ith the increase of the number of generations. > Sugge!'>tions: I rom (I) and (2), I• . ~- ~·U '- ( • I' 1 • I', • I'll ) (.\)- !WicJII +\alx1 + 0 2·'2 · + ...... +owxlf · s~nsiti\ it) or equation (3 ): [ • .,, ( I' I' '(.\)"-./umou+ 0 1-'1 +a~x~' !- ..... . tF (x ) w 1 1 s· ~ = cl p,x1 = , I I'.\ I i"if-'(.y) _ . If'! IJ _ <' ~ ll!f>!.\! - .) , CY, i'Flx) . ll' 2000 '"" 19, 359, 000 and the popnh1tion ~rowth n1te was lA as a pereent>1ge. Compared to the "'lne of popu I at ion , the latter value is nc~ligihlc. llut, ;iuec it i> considered that each and "co·y fac!o r ('a riablcs of the fi t UN fuuct ion) h >~> c!JIHII sensitivity , any of those .tgu rc'> can not be neglected. rhe mcthodolog)' "Input-scaling" could be practiced at the stage of GA tra ining. When consider the whole .,et of data from yr. 198-l til\)' r. 2001, the smallest value is 0.032 (unit price of electricity - Indus. and comm. in US$ in yr. 1986) and the largest value is 19, 359, 000 (population in yr 2000). All the other li~urcs \ary with in this range. So, input scaling is done by mapping 0.032 to 0 and 19, 359, 000 to 1 as shown in the following tigure. l)calcd do" n \a lues • 0 .. .... .... ~-· ---~-~------ : 0.032 19, 359. ~000 " .,.. Actual values • If the number of generations could have been increased to a higher value (e.g_; {' (, () , 000), more clear idea about the convergen<;_e. /~ Pagc61 of88 CHAPTER 5 Forecasting Model for Electrical 1~nergy Demand of Sri Lanka :;.J Pr<.'paring the for<.'casling ;\lod<.'l When consider the Genetic Algorithm. it has been trained for a particular set or data. ~~~ the accurucy or the output is high with in the state-space or the input J'ta. When the 111put tl:!ta mo\'es beyond the stat ·space consitlcred. gradually th~ accur.JCy or the output IL'duccs. As discussed in chapter 4. with the \<.tlucs obtained at the Bt.:st Fit condition after 20,000 number of generations was used in preparing the model for Electrical Energy I kmand Forecasting. I rom (2}. /· ( \)- (0.7978f") t (- 1 .18()24 X .\'1 1 ·'' l.;s )+ (0. 742877 X X 2 OOili'J<)l )+ (0.128874 X X 1 1 570 :) ·t· (-()<)2·'89. "l:'ll I) (l o~,7?9 . Ol'IJ~) .... (-19865 1 .·l?76'l~) . (1 ")'i7J74 .o~~~!Jll•) . t > .\ 4 1 , .• l. _ X \~ • . -+X .\1, f .. _ X.\ J + (- 0_44()?,3 X \ 111 O~OI1\ )+ (-1.64072 X .\'1- loli'IJ) "'~ ................. (3) ; I he tlescription ol'tht.: \'ari..:~hles in equation (4) ts given in the f-'igurc 5.1. v ~ / Pagl! 62 of'88 -( Raint:tll in catchment areas in n1111 ~ .\J I (X I 000) -'".' GOP per capita in l S S ( x I 00) -.. X_; --,. Population ( x I 07 ) Y.; --,. Population grtn'vth rate (as a 0/o) '"' ,... A veragc l 1 ~ S "alue in rupees ( x I 0) X (I ... Domestic consumer accounts (xI 0 5) x· .. A \C. unit price or dum. consump. in -... lJ~$(x 10~) \" s ... A vc. unit pri~.:e of Indus. & Com. COil'\lltnp l JS $ (x I 0"2) j X•J ... ... I\ vc. unit price of other con sump . accounts lJS $ (x I o·:!) ) v Figure 5.1: Details of the factors con::.idcrcd in the model llctK<:. \\'ith the available real time data for the abo\ c ntctors in year 2002 & 2003. the forecasted cb:tricity demand f(v corresponding years u::.ing (4): 5.2 1-.kctrical energy demand for year 2002 = 7. II 091 T\Vh \"kctrical energy demand for year 2003 = 7.17532 T\Vh Error with the forecasted data --/ "' ·" .. ' --·- I JT\ll. \\ ith the forecasted demand with respect to the actual demand could be dcfi1lcd as, ( actual dem . - rorccaskd dcm . u error= - - x I OO~o actual demand Page (>3 of 88 l "nr )Car 2002: ( (>.75559-7.1 1091 )xi00% - 5.7:)673 1% 6.75559 I or \car 2003: (7 61 '1 - 7 . . 17532 ) lOOo' _ _ ,1')96-l0 u . - X 10- - ' ·-- 7.612 5.-t \"alidity of the for·ecasting l\lodcl j /\ctual data used to Validity check Forecast for next train the model ------... 2 years 19X4 200 I 2003 I "hi: moc.kl has been trained \\ ith 18) cars (from year 1984 till year 200 I) of actual data. Then the \alidity of the results \\as checked for the next 2 years. The table shown 111 I able 5.1 presents a comparison bet\\Cen the accurac) of the CiA model forecast and th~: I nne trend forecast (done b) the C r.B) for these t \\ o years ... ~ " Table 5.1 Comparison of the GA model forecast and the Time trend forecast (done by the CJ<:B) for these two years. Yea 200 200 200 200 200 :- j 3 I t ) ' 5 I i Forecasted Demand error % 7 110906 -5.25319 7 175318 5 .73675 7 .668869 4 65164 7 836188 8.069451 . l Actual Demand Time Trend •• • 1- (TWh) - Forecasted demnad --~rror % by the CEB " ~ - - ___L_ 6 756 7.381 -9251036 l 7 612 8.106 -6 489753 8 043 8.889 -10 518463 8.889 9 748 Page 64. of 88 l:lectricity Demand of Sri Lanka with the model forecasted data is appeared in figure 5.2. 8 000000 7 000000 .J::; ~ 6 000000 1- c 'U 5.000000 c nl ~ 4 000000 Cl j >-1i 3 000000 ·;: ..... u ~ 2.000000 w 1.000000 0 000000 1980 1985 1990 1995 2000 2005 Year -+--Actual Elactricity Demand Model Forecasted Demand Ftgun.: .:'.2: Energy Jcmand \S. Year (actttJI data and the modd forecasted data) .:;_s l( lectricity Demand forecast .. ~ " In the process of forecasting Electrical Energy Demand of Sri Lanka, forecasted data or each factor under consideration (the factors that arc been considered in designing_)l~e lim.:casting model) are needed. ~ourccs of !orccasted data: ~ ..... .,- ~ • Cl.--.B domestic consumer accounts / • By doing a time trend analysis to obtain the rest of the data by myself. \ppcndi.\ II shows the forecasted d:lta under each factor. Page 65 of 88 I he fon.:castcd electrical energy demand \vilh the above data is shown in Figure 5.3. 12 -..c ~ 10 ..... - -o c 8 ra E Cl> 0 >. 6 .... u ·c .... u 4 Cl> w 2 0 : .: . . ---~-- ... ~-· ···· · · .- 2002 2003 2004 Year 2005 Forecasted Demand • Time Trend Forecast . --. - . j 2006 _.._ Actual Demand I l _!.!liiC 5.J: Uectrical Encrg) Demand li.m.:~ast dom! with the moJel Shorr 'I crm Considering a 55% Load !·actor, The possible Peak 1-lcctricity Demand could be pn:senteJ as shO\\n in Table 5.2. ..~ " l'al>lc 5.2 The poss ible Peak E lect ricity Demand cons idering a 55(Yc, Load Factor Forecasted Year Demand (TWh) Load Factor --- 2004 7.668869 55% 2005 7.836188 55% 2006 8 069451 55% Max1mum possible peak demand (MW) - 1591 .712121 1626.440017 1674 854919 I ~·-.,r ~ ; '. /~ lhe I igurc 5.1 shO\\S the load Clii'\C or an an~ragc da) or the Sri Lankan Power S) stL'Ill I he Peak demand occurred around 20 00 hours and it was 1604 M \V. This tk:111anJ could be met \\ ith the forecast done "ith the < 11-\ based model: i -~- 1626.44 M \\'. Page 66 of 88 1800 1600 1400 ~ 1200 ::i! ~ "0 c: nl 1000 E Q) 0 .~ 800 (.) ·;: .... (.) 600 Q) w 400 200 0 -1 00 4.00 900 14 00 19.00 24 00 Time Figure 5.4: Load curve or \ri l.anka on I ~~ June 2005 .. ~ ( 'ommcnt on Section 5.-t and 5.5: Accord ing to the above validity checl< as nell as the forecast, it is shown there is a higher accuracy with the forecasted data when compared with the actual avaKablc data. > ..,.._ If the GA could ha\'e been trained with actual data up to t-he year \)'e know, "ith a larger number of ~enerations (e.g. 60, 000), more accurate short term demand forecasting cou ld he obtained. Page 67 of 88 5.(, Conclusion rhi s tlH:sis ha~ described a 110\ cl concept of forecasting ckctrical en erg) demand or \ri Lan ka. l he e!Tectivenes~ of the proposed methodolog) based on C1enctic \lgorithms !< i :\ ) opt imi/alion '"as demonstrated b) the results. I his (i/\ based electrici ty demand forecasting model has several advantages. It was llllKie lcd considering 9 major f~1ctors that could affect the electric ity demand or Sri I anka. lhcy are GDP. Population. Population gro'' th rate. 1\ \'Cragc annual rain-fall in catchment areas. A \erage L S S \<.due, Domestic consumer accounts. l).~·erage unit price . of donJc-.,tic clcctricity consumption (in l iS $), AH:rage unit price of industrial and collllllcrcial elect. consumption (in lJS $), /\ vcragc unit price of other elect. consumption accounts ( in US $). In thi s model TIMK has not been considered as a factor, s ince the electric ity demand of the country i-., dominated by the several other components ( Discusscd in Chapter 3) hut not time. l he applicability of CJ ,\ for the problem and the validit) fnr the clcctricit) demand of the h ucca-.,ting :V1odel have been verilied experimental ly. This GA based model would support the short term electricity demand forecasting of a given system. \lore accurately fo recasted data for ea<.:h fa<.: tor \\ ould predict the Electrical Energy I kmantl \\' ith a larger accurac) . In the prediction of figures of.:~ch factor. support of the < il'IH:tic Algorithms is high!) rccomlnendcd. .. To improve the accuntcy of the results, ... .. • Input Scaling could be practiced at the stage of GA tra ining. " • ~umber of genera tions could be increased to a range such as 60,000 or more. • Usc of more improved programs (GAs) with less processing time etc. COt\TINl fEI) ... . .. . Page 68 of 88 • lise of computers with higher memory capacity • l lsc of more speedy computers • In ot·dcr to run such programs specially committed computers for the wot·k concerned an~ essential. 19X4 Train the G.\ \\ith available data Forl!ca:;t for the ne>.t 3 years "ith thl! model based on GA outputs Year o f' latest data avail able .. ~ ... l .· ..:...__\. ...... y J<- ~ / Page 69 of 88 IH H'RFI\CES Ill hhted h\. r Ekctriclt) hup· \\\\\\ .terrin.org,'dr\rslon lrcgdl\ ldol·s flU pdf (lOth Janu,rry 2005). Short f'l'llll Fb:tricit} Demand forecast of Sn I anka. 2002. System planning branch, CEI3. hi\ 11onment & generation plannrng hr.111ch. Cc; I on l:lt•ctncrty Uoard. "Dcmaml forccastrng for 200 1 ( ieneratron Lxpansion Planmng Studres. 2001 ", ppl--1 J I he annual n:por1s of the Centrall3ank of Sn I anka. ( olomho. Sn I anka. I %4 -200-1. rq '''"'""' Repor1s of the Cc.:ylon Llcctnclt)' Bo.ud. 19:-1·1 200-1. I I I d.rk "ir;amhalaprt rya. "Crisrs 111 the Lkctrit:lt) Sc~.:tor and Solutions". Resource Management \~~, S1 1 l.anka. 200 I. I d.tk Sryambalapitiya, "'I he Real Problem in the r: kctncrty Sector", Rc.:~ourcc :\1anagement \ s~(ll. ~ '0 121 ', I}' ''..) '" R.11 n f:.dl data in catchment areas S11 Lanka mctcorologrcal Department. St.rtl\trcal tk-partmcnt data \ Ssm(,e~ . E. Costa. "Parametnl· '>tudy Ill l:11h.IIICC ( icnc!Jc ,\ lgo11thm' s l'cii(Hmar~~:e \\hen Usmg l r .lll~!illmatron ", Proceedings of the Genctrc and b ollltrOII,II y Computation Confi:rcncc (igns", Sprin~cr - Vc.:rlug I ondon l 1111rtcd. 19~>9. pp 1-43. . . ' ,\ ( 'hrppl'rficld. P.Fiemmg. I f.Pohlhcim, c .ronM.'\::1, "( icnetrc Algon thm roo I 80\ for US<.: with \lathlah 0 Vers1on 1.2, User's (lurdc". lkpartmt·nt of 1\u((lmati<.: Control and Engineerin~:,.,..; l Jnl\ er~11y of Sheffield. .,._ " K K. Y.\\'.I'ncra. "An [\'aluation of the TIL'Itds 1n the Energy '>ector and Potential for Dev~loprng f{ l'nl·wable Energy". Sri Lanka Fconomic Assocratron, Colombo 0.\, 1993. / l(, \ ( hrpperlield. "Introduction to GL·nctre Algonthms", in "Genetic t\lgorithms 111 Engineenng s~St l'lll-." Edrtcd hy. A.:\I.S./al7ala. P.J Fll'ming, The lnstrtution of Ekctncal l·ngincers. l. K, I 997. np I-SO http .• , \\" ' ' .n·h.l k · gcncrationlren·nue hmly.hrm ( 03r J (kt. 2005 ). Page 70 of 88 \I'PENDI X I htncss ':1lucs obta ined" ith different par·amcte r setti ngs ( ILt-.,cd on data from year 1984 till year 2000) Rewlts Jl•hen Mutation Probabilizl', Pm = 0. 7*250/Und ~1:'\1) = 100: - 'llllllbcr or indi\ iduals per :,ubpopulations - :Vla.\imum >-lumber or generations t\ 1.\ \GLN- I 000; lrli:\P - 0.9: -Generation gap. how many new individuals arc created 1\ \' ,\R = 1 S: - (iencration gap. how many llC\\ individuals arc created I'RI Cl- 20; I iddD - Prccbion of binary representation -Build field descriptor [ 161 l· icldf) = lrep (lPRECIJ, II, NV/\RJ); ... -~ -; -~ -i -; -; -i -; -~ -~ -; -; -~ -; -~ -~ -; -~ ~::} rep(! 1: 0; I ;IJ. [1.1'\VARI)I: Best= 1 0299 400 j l.'pper and IO\\cr limits (range) or o\t and p,,. .. ~ " I · r~ ~ ; ' . 600 generation ~BOO 1000 1200 ••• Rcsults- Columns I through 6 -1.5251 -1.1937 0.9246 0.7158 -(>.0089 -0.8835 Columns 7 through 12 1.8537 0.1487 -1.5434 -1.007(> 0.4868 0.3781 Columns 13 through 18 -I. 7182 - 1.8608 -1.1 003 -(UU56 -0.4940 0.437 · ... ~ .. .;' ~ / l'agc 71 orss Nl ;\1 1) 100; r--.1A Xl · rep (j I : 0: I : I ] . [ I, 7\ VA R [) I: 15 ~----~------~-----.------.------.------. Best = 0 96224 I 10 (f) (f) Q) c:: LJ.. ill (LJ !Il I sr 0 ~----~------~------~----~------~----~ 0 2000 I ~ L'" ul t :-. < olumns I th rough 6 1.0076 0.5993 -1.4 192 l 1 )l umns 7 through 12 () \ 17·l 0.5689 -1.7132 l tllu mns 1.) through 18 4000 6000 generation 8000 - 1.7846 -o.sg - .> -->- - .) - .) -.> -.>- -.> -.> -.> -.>- -.) -.> -.> :... ppcr anl O\\'Cr 11111ts ) 33 3333 33 311333 3 1J: ... (rangc)of'a,1 andJ>r-.1. rcp(fi:O: I :IJ.Il.NvARj)l: 30~----~-----.------.------.------.-----~ Best= 1 2667 l 25 20 (/) (/) w c:: Li: 15 Ui w [I) 10 0 1000 2000 3000 4000 5000 generation --/ .. 6000 R~.·~(dts ( 'olumns I through 6 2.4247 -2.4113 -2.8759 -2.4226 2.(H02 - I .2192 ('plumns 7 through 12 -0.4186 -0.9009 0.8850 0.893 1 0.7326 -0.7946 < 'olumn<> 13 through 18 -0 7 -l88 -2.2293 -1 .8770 -0.1972 0.0252 I . 7023 , ·""· Page 73 of 88 '\INJ) 100; \1.\XGEr\-" 10000; (i(l •\P 0.9; '\\' \R = 18: PRf ·CJ 20: 1-lt:ldD ~[rep (lPRECI]. [ L NVARJ): ... -3 -J -3 -3 -3 -3 -3 -3 -3 -3 -3 -J -3 -J -3 -3 -3 -3: ... J 3 3 3 J 3 3 3 3 3 3 3 J 3 3 3 3 3; ... rcp([I:O; I :1),[1,\VARJ)j; 80 Best...;, 0 98872 70 60 50 (/) (/) (l) c ti: 40 u=; (l) (IJ 30 20 10 J ,_ ·r- -_r . 0 2000 4000 6000 BODO generation Rt:sult~- Columns I through 6 O.M~ 12 0.9254 0.9343 0.6422 -2A843 -0.308 1 ( ' olumn~ 7 through 12 02420 2.1772 1.8777 0.5213 -2.7037 -0.2(>58 ( 'olumns 13 through 18 -O.O~J I 0.8562 -2.9682 -1.4105 -2.(> 174 · 2.1406 j "-.1'0000 12000 . . • ..... .; , . ;". ·~ /~ Page 74 of 88 Remits u·hen Mutation Probabili~J ', Pm = 0. 7*400/Lind '\IND 100: \lA '\GFl\- I 0000: (,(j\P 0.9: '\;\. \R IS: I'Rl Cl - 20: I il·ldD- [rep (fPRECI), [1. 7'\VAR]): ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 : ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 ; ... rep ([I: 0; I :I]. (1. N\'AR)l]: 30.-----.-----~------~ Best = 0.9447 25 20 (f) (f) Q) c u:: 15 iii Q) CD 10 5 ob 0 2000 4000 6000 8000 generation Rc.:sults ( qlumns I through 6 () 843<) -1.3774 1.1086 0.740(> O.OUQ 1.1 s 19 Columns 7 through 12 - 1.1449 -0.3957 0.3422 0.3424 -0.7233 -0.8952 Columns 13 through 18 -0.1905 -1.9047 0.4735 -0.7872 -0.1141 -1.5473 f .,.1, ... 10000 12000 ;' . . . .. ; ~ .. ·- v /~ Page 75 of 88 Newlts wit en Mutation Probability , Pm 0. 7'~500/Li/1(1 1\1'\[) 100; ;\L\\:GEN = I 0000: (1(1.\P 0.9; :--.:\1 \R 18: P!: l C I- 20; Ftci<.JD =[rep ([PRECI) . [1. ?\VAR]); ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 2 2 -2 -2 -2 ; ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 ; ... rep ([L 0: I :I]. [I. '\\'ARl)); ·7 Best= 0 69783 I 6 5 ~ 4 ru c u.. ]j 3 m 2 0 0 2000 4000 6000 8000 generation Results Columns I through 6 - 1.(>527 0.:\665 1.9263 0.33 14 -1.0 139 0.7109 < olumns 7 through 12 I 2844 -0.8623 0.9255 0.6094 -0.1223 1.52<)8 Co I umns IJ through 18 () 5933 0.3651 1.8038 0.01J8 -J.1J)2J -J.OJ7J ~ #' .. / "' 10000 12000 . _;' ·~ :.*- vl ~ / Page 76 of 88 Results II' !ten Mutation Probabili~l'. Pm 0. 7'~ I 000/Um/ \!lND ~ I 00; \lt\XCiEN I 0000; < iCi \P = 0.9; '\\ \R- IR: I'RI (I 20; held[) lrcp (LPRECI), [1,1\VAR]); ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 2 2 ; ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 ; ... rep ([1; 0: I ;1]. [l. :\VAR])]: 20 18 16 14 (f) 12 (f) Q) c J:: 10 (j) Q) CD 8 6 :~ 0 2000 4000 l{l'"lllts Columns I through(> Best = 0.88647 6000 generation 8000 -I.HJ95 -0.42 19 1.7896 0.0257 -0.434 I -0.5X39 Ctllumns 7 through 12 1-.92 I 8 -0.3233 1.0436 0.3747 1.0628 - I .53 I 2 < 'olumns I 3 th rough 18 () ~()():) 0.5234 -0.3 I 24 -1.0164 I .OS<> I -0.101(> -7 "' 10000 12000 ...... y/ , ; ' ·~ ~ / Page 77 of' 88 R('\11/ts ll 'ilen Mutation Probability, Pm 0. 7*500/Li/1(/ '\ l \JL) I 00; \1\XGE\l= 10000; <,(,\P-0.9; '\\'AR-18: I'Rl Cl- 40; Fici!D- [rep ([PRECI]. [1. :\\'AR]); ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 2 -2 -2 -2 -2 -2 ; ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 ; ... rep ( [ I ; 0; 1 : I ) . [ I , :\VA R])]; 7 I Best = 0 87006 I .6 5 r.n ~ 4 c u.. ]j 3 m 2 0 0 2000 4000 6000 8000 generation Results · ( 'olumns I through 6 -0.7057 0.1503 -0.3770 -1.6977 - 1.6()95 1.6845 < olumns 7 through 12 . I 2:' I<) 0.6 184 1.8636 -0.1017 O.J 175 -0.":> 775 Columns 13 through 18 I 5402 0.5034 -0.1287 -1.5052 0.400-1 0.:'1004 --!.r " 10000 12000 ; • .,..; :..._ v ~ / Page 78 of 88 Remits wlteu ilJutatiou Probabili(J', Pm = 0. 7*500/ Liud :"Jl~[) ~ 100; :'\I \ '\ < i I· '\J 20000; (j( I \P 0.9; ~\ \R- I~; I'RI ( l 20; hcldD- [rcp((PRECJ],( I. l'\\' AR]); ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 2 -2 -2 -2 2 -2 -2 -2 ; ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 : ... rep ( [1: 0: I :I]. [I. 0:\. A R])]: 14 lJcst = 0 70225 12 10 (/) (/) 8 Q) c i.L iii 6 Q) CD 4 2 0 0 0.5 1.5 generat1on Results Columns I through(> -O..f052 1.7932 I .4570 (U500 1.9972 -1.5002 ( \llumns 7 through I 2 1.0014 -1.3785 -1.2461 < (;lumn~ 13 through 18 -0.4979 Cl.7 '42 -I.W90 1.230 I 0.4015 -0.6925 -1.9561 - I. ()<)4 7 -I. 8-l97 l --~ ... 2 2.5 4 X 10 ...... .., , ... Page 79 of"88 1\ IND 200; :VL\X<.1EN 20000; <• 170 -1.4907 j 12000 . . . ~-. .... .., ,; / PagcSiofSS '\IND 400; :\1:\:-\GEN -- 20000; (i(ii\P 0.9; :-.J\· \R- IS; PRH'I 20; hddD !rcp((PRECI].[I, NYARJ); ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 : ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 : ... rcp([l; 0: I :I). [1, N\'AR))); 5 Best'= 0 67783 45 4 3.5 (I) (I) 3 Ql c u_ u; 2 5 Ql (!) 2 1 5 1 o5t 0 05 generation l{csul ts < ·o lumns I through 6 I. "363 -0.1313 I .061 I 0.5355 -0.307R Columns 7 through 12 1.5 2 .. ~ "' 1.9737 1.5057 -0.7148 -1.9642 -1.5308 1.1 X88 -0.7728 ( 'o lumns 13 through 18 () 7418 0.4505 -1.4299 -1.8278 -0.473(> -1.6152 I 2.5 X 104 .... . ..... , Page 82 of 88 /~ '\ IND 400; \L\ \CJH\ 20000; GG•\P 0.9; '\\\R·IS; PRI·C I- 20: heldD -· [rcp([PRECI),[ I, NVAR]): ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 : ... 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 : ... rep([ I: 0: I :I]. [I,\ \':\R])J: 0.5 1 1 5 generation R-.:sults Columns I through 6 (un2s -0.0338 0.321 <) 0.5761 -1.7(>X(> Columns 7 through 12 0.4S82 1.3843 I. I 97 I o.33m 0.3<>77 - 1.4<)5<) Columns 13 through 18 - ,2 ""' l.lllJO I. 75~7 0.3<>54 -l.U3o -1.(>~22 0.1 (>0) 0.840<> Rc:,u/ts when Jlutation Probability, Pm = 0 .... *500/Liml 2.5 X 104 . ; '. . . . ..... "" ; / Page 83 of 88 NIND- 500; M 1\XGEN- 5000; (j():\J> 0.9: ~\ \R 12: PRI("I-20: IJcldD [rcp([PREC'I],[I, 1\Vt\Rl): ... -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 -2 , ... 2 2 2 2 2 2 2 2 2 2 2 2 ; ... rep([ 1: 0: I :I], [ L :\VARJ>l; 1 5 I I 14r 1 3 en ~ 1.2 c: l.L u; ~ 1 1 1 09 Best= 0 80135 j - - ·----, I ----..,__ I I I 0 8 ~ 1 0~0 ?n~n ":Jnnn mM 20~~ l{c..,llfts - ( ·nlumns I through 7 3000 gen.erat ion 4000 5000 1 -A.r 1.2837 -1.()859 -1.0000 -1.7500 0.4·-l)g 0 . ."1750 - 1.2656 Columns 8 through 12 -0.7500 0.1 602 0.5845 1 .3~20 0.4175 6000 .#'·· ·~ ... _ ., Page 84 of 88 ,/' \PPF.NI>IX II Data used for the forecast l·igurc 2.2: The electrical energy demand or Sri Lanka from year I %5 till year 2003. Year ' Electricity Demand (TWh) ' - 1Cl84 I 2.250083 1985 2.462867 1986 2.642101 1987 2.692351 I I 1988 2.784288 1989 2.843976 1990 3.1 33769 1991 3.354094 ' 1992 3.509674 1993 3.952684 1994 4.338150 1995 4.756927 1996 4.338189 1997 4.872005 1998 5.517904 1999 6.027877 2000 5.258206 I 1 2001 6.626548 I ".:" 2002 6 755590 2003 I 7.612 -- -- - ,.,· . .. ·- ., Page 85 of 88 ~ lhc z coloured ligures fi gures used in dc~ i gnin g the fo recasting model. I he I I co loured ligures ligures used in f(H·ccast ing the Electric ity demand. l'lu.· figun.·s U!ied in fo recasti ng th t• Elect r ical energy dema nd of S ri Lanka -- ~-- - --, Ave j I I Unit pnce I Untt Unit Ramfal 1 of elect pnce pnce of lin GOP I I (USS~W of elect. catch per Popu Popu A US h) Dome. elect (USS /k cap1ta (X growth ve Year ' ment 1 (o;. 1 S value I Dome Consu (USS I Wh) area (in US 1000) rae o 1 X X accounts X. kWh) I Other ~v., $) X I '>: 1 ~~"~"' X X. I I I I I Com -r- 1 · X 15.603 T~2- --~ - -1984 25 48 0 046 295. ~4 0.078 I o 046 2606 51352 1985 2368 6 344 15.841 1 5 27 21 I o 041 I 329, 965 I 0 068 I 0 042 11986 2304 6 362 16, 127 1 8 28 07 1 o o37 370, 048 0 .032 1 o 04 1987 2019.4 368 16,373 1 5 29 5!>' 0 037 404, 962 0.034 1 o 042 1988 I 2240 1 385 16,599 1.4 31 90 0 049 450,431 0 074 j 0 047 1989 I 2329 4 374 16.825 1 3 36.33 0 042 495 932 l 0 061 1 0 039 1990 I 2160 5 437 17,017 1 1 40 09 0 047 628, 741 1 0066 I o 037 1991 I 2110.3 469 17,267 1 5 41 45 I o 051 751.614 1 o o? I 0 036 1992 2041 7 557 17,426 1 0 44 18 0 049 917,319 I 0 076 1 o 04 1993 I 2404 o 588 17,646 I 1 2 1 51.61 0 045 1,089, 287 1 a 075 1 o 037 I 2213 7 I I 1994 656 17 891 1 4 I 51 87 0 049 1, 222, 124 I 0096 1 o 047 1995 I 2185.7 I 719 18,136 1 4 49 92 0 045 l' 322.087 I o 046 . 0093 1996 2132 4 I 759 18,336 1 1 57 58 1 o 046 ) 466,815 1 o ogo 0 045 199i 2447 7 1 853 18.567 1 2 59 85 0.047 1, 611, 102 I o 091 I o 044 i I o 044 1998 21 11.5,879 18,774 1 3 70 39 0 044 1. 781, 388 I 0.088 1999 22130 863 19,043 1.5 75 78 0 040 1, 981 , 69 1 0 082 0.040 . y , 2000 I 2050.5 I 899 19,359 1 4 89 36 0 034 2, 191, 301 0 082 0 038 2001 I 1972 4 I 841 118.732 1 4 95 66 0 041 2. 364, 853 0 079 0 .042 I I ~·· ' 2002 I I ...... I I v I j j I "' 2003 I ,.,. 2004 2005 2006 2007 2008 I I I I I I . I' I ) • j _ I I I I I f&L Figure 5.2: Energy demand vs. Year (actual data an<.! the model forecasted data) :\ctual Data Y - 1- Electricity ear Demand 1984 l 2.250083- 1985 2.462867 'l986 2.642101 1987 2.692351 1988 2.784288 1989 I 2.843976 1990 3.133769 1991 1992 1993 1994 3.354094 3.509674 3 .952684 4.338150 1995 I 4.756927 1996 4.338189 1997 4.872005 1998 5.517904 1999 6.027877 2000 5.258206 2001 6.626548 2002 6. 755sgo- 1 2003 _J_ 8.043 __j Model forecasted Data --r --y Electricity I car Demand ~ 2002 J 7.110906-2003 7.175318 I A~ " , 6 . ... ~ ..... v /~ Page 87 of 88 Figme (>.5: Loa<..! curve of Sri Lankan Power System on 01'1 June 2005. (Readings taken by the System Control Division, at 30 minute intervals) lime I Generation( \.1\V) I Time General ion( M \V) - O.JO 775 12.JO 1082 I . 00 73'D I I 3.00 1069 I 1..30 71~ I IJ.JO IOS1 I 2 ()() 70(> I 14.00 1097 2.30 701 I 14.30 1120 3.00 694 15.00 1114 .\30 693 15.30 1115 4.00 700 16.00 i 1130 4.30 723 16.30 981 5.00 770 17.00 1027 5.30 8<)4 17.30 1003 (>.00 1 on 18.00 982 (>.J() IOR2 18.30 1025 7.00 C) 57 19.00 1229. 7.30 885 19.30 1575 8.00 914 20.00 1604 ~.JO 979 20.30 1575 <). ()() 1045 21.00 1510 9.30 I 1073 21.30 1423 I 0.00 I 097 22.00 1252 I