基于灰色理論的短期電力負荷預測(精)

1、設計（論文）內(nèi)容及礙求：設計內(nèi)容：L解EMS條統(tǒng)柑關期識2.確定預測冃標.搜集與整理資料3.對電力系統(tǒng)垢期負荷預測進行較為系統(tǒng)的研究4.分析資料，選擇預測巧法3 3確定矩期負商預測方法6.建立短期負荷預測模里7.對短期負荷預測進行仿真實驗研究&進行預測分析二、 設訃要求：L翻譯該課懸相關英丈論丈簫2.設計說明昔一份（含中英文摘要*正文、程序清單）三. 蔘考資料上I.管理系統(tǒng)2+電力條統(tǒng)門動化雪宵關電力條統(tǒng)負荷預測方而的參與文獻工仃關MATLAB/S1MULINK f方4方商的牧材及鯊科4.神經(jīng)網(wǎng)絡技術5.能擰制理論6+電力系統(tǒng)短期負荷預測指導教師】年H日本科生畢業(yè)設計（論文）開題報告設

2、計（論文題目廉于灰色理論的垢期電力負倚預測設計（論文）題目來源自選題目設計（論文）題目類塑理論設計起止時間2007.12.132008.6一、設計（論文）依據(jù)及研丸直義，依據(jù)： 電力負荷俺酒對于保證電力工業(yè)的倣康發(fā)昶， 乃至整個田民經(jīng)濟的發(fā)展均有召十分呃 耍的怠義.仮俏預測依其運用鞭域可分為；運轉規(guī)劃、電源開發(fā)及電力系煉規(guī)劃袴三 爬不同應用泱域所需負荷做構之內(nèi)容亦不盡郴同.因此.負荷張河的模式從其所 便用的救學模式q公式.訴期齊電力串業(yè)不問秤景可環(huán)竝的條件而nun大的建并 愆義：準確的負簡預測.可以避免經(jīng)濟合珅的安掛電網(wǎng)內(nèi)部發(fā)電機組的啟停，保持 電網(wǎng)運行的安全也運性.曲少不必要的庭轉儲備容碗，

3、仟理安排機組檢修計 劃.保證社會的正常生產(chǎn)和生活.冇效地降低發(fā)電成本.捉髙經(jīng)濟效益和社 會效益.二、設計（論文）主晏研完的內(nèi)容、預期目標：（技術方案、路線）內(nèi)釈k進行系統(tǒng)分析2、 隹立灰色系統(tǒng)模樂GM模型即灰色模型（GREY MODEL）,一般來說.冬模是用原始的數(shù)堀序列建立 左分方用：灰色糸統(tǒng)建仗則足川原始數(shù)據(jù)仔列件牛成數(shù)后建工做分程.由于糸統(tǒng) 被噪音汚染后.所以原始數(shù)據(jù)序列呈現(xiàn)出離亂的怙況，這種離乩的數(shù)列也是-種灰 色數(shù)列，或？5灰色過程.對灰色過程雄立模型，便成為灰色模型.3、運用灰色理論進行負荷傾測灰色系統(tǒng)理論研丸的是貧信總卜建模.提供了貧信息卜解決系統(tǒng)河越的新途彳盒 它把切隨機過程看

4、作定金淀范用內(nèi)變化的足叮時間仔關的尿色過程.對灰色磺 不是從統(tǒng)計規(guī)律的角度應用大樣本進行研克耐是采用數(shù)據(jù)生成的方法將雜亂兄&的原始數(shù)據(jù)整理成規(guī)律性強的生成序列再作研究.4.預測i杲差分析任對試算結果逬行統(tǒng)計分析中發(fā)現(xiàn)：fiiiWH負倚伯時如果同時出現(xiàn)氣候溫度 突變的悄況.預測準確率也會卜降對此.我們決定：根據(jù)氣候溫度的夫變程度分 岀幾個不同的調(diào)整權取.溫度以289為分界，低F28X?每相套代為-檔,髙r-28*C毎相芝2C為一檔.預期II杯：2007.12完成翻譯2008.3收集負料2008.5建楔2008.6編程三、設計（論文）的研兀點及難點：權點：龍丄灰色模住及其改進模熨堆點：灰色

5、松唱的數(shù)學妊模及其MATLAB程序的編行四、設計（論文）研丸方法及步驟（進度安排h設計研究方法：以定性分析為主步fflh丨、確定負荷預測I的.腳訂預測計劃2、 搜孑、聲理、分析資料3、 妊立預測楔型、運用MATLAB軟fl編程及仿負4、 確足預測結果，分析觀差5 5、編寫預測報告耳垂行設計（論文）所1條怡1電力系統(tǒng)垃用Ml荷預測樣本故貝；（某市2003年II月電力負荷實際數(shù)據(jù)、該市2003年11月天氣情況的數(shù)據(jù)）2 有關負荷預測和灰色理論的期刊和書籍、MATLAB軟件簽舗年 月 nLOAD FORECASTINGEugene A. Fein bergState Uni versity of N

6、ew York, Sto ny BrookEuge ne.Fe in bergs un Dora Gen ethliouState Uni versity of New York, Sto ny Brook dge nethlams.s un Abstract Load forecast ing is vitally importa nt for the electric in dustry in the deregulatedeconom y. It has many applicatio ns in clud ing en ergy purchas ing an

7、d gen erati on, I oadswitch ing, con tract evaluati on, and in frastructure developme nt. A large variety ofmathematical methods have bee n developed for load forecast ing. In this chapter wediscuss various approaches to load forecasti ng.Keywords: Load, forecasting, statistics, regression, artifici

8、al intelligenee.1. In troduct ionAccurate models for electric power load forecast ing are esse ntial to the operati on andplanning of a utility company. Load forecasting helps an electric utility to make importa ntdecisi ons in cludi ng decisi ons on purchas ing and gen erati ng electric power, load

9、 switching, and in frastructure developme nt. Load forecasts are extremely importa nt for en ergysuppliers,ISOs, finan cial in stituti ons, and other participa nts in electric en ergy gen eration, tran smissi on, distributi on, and markets.Load forecasts can be divided in to threecategories: short-t

10、erm forecasts which are usually from one hour to one week, mediumforecasts which are usually from a week to a year, and Ion g-term forecasts which are Ionger tha n a year. The forecasts for differe nt time horiz ons are importa nt for differe ntoperati ons within a utility compa ny. The n atures of

11、these forecasts are differe nt as well.For example, for a particular region, it is possible to predict the next day load with anaccuracy ofapproximately 1-3%. However, it is impossible to predict the next year peakload with the similar accuracy since accurate Ion g-term weather forecasts are notavai

12、lable. For the n ext year peak forecast, it is possible to provide the probabilitydistribution of the load based on historical weather observations. It is also possible,according to the industry practice,to predict the so-called weather normalized load, whichwould take place for average annual peak

13、weather con diti ons or worse tha n averagepeak weather con diti ons for a give n area.Weather no rmalized load is the load calculated for the so-called no rmal weather con ditions which arethe average of the weather characteristics for the peak historical loads overa certain period of time. The dur

14、ation of this period varies from one utility to ano ther. Mostcompa nies take the last 25-30 years of data.Load forecast ing has always bee n important for pla nning and operati onal decisi on con ducted by utility compa ni es. However, withthe deregulati on of the en ergy in dustries, load forecast

15、i ng is eve n more importa nt.Withsupply and dema nd fluctuat ing and the cha nges of weather con diti ons and en ergyprices in creas ing by a factor of ten or more duri ng peaksituations, load forecasting is vitally important for utilities. Short-term load forecasting canhelp to estimate load flows

16、 and to make decisions that can prevent overloading. Timelyimplementations of such decisions lead to the improvement of network reliability and to thereduced occurre nces of equipme nt failures and blackouts. Load forecast ing is alsoimporta nt for con tract evaluati ons and evaluati ons of various

17、sophisticated finan cialproducts on en ergy pric ing offered by the market.In the deregulated economy, decisi ons on capital expe nditures based on Ion g-termforecasti ng are also more importa nt tha n in a non-deregulated economy whe n rate increases could be justified by capital expe nditure proje

18、cts.Most forecasting methods use statistical techniques or artificial intelligence algorithmssuch as regressi on, n eural n etworks, fuzzy logic, and expert systems. Two of themethods, so-called en d-use and econo metric approach are broadly used for medium-and longterm forecast ing. Avariety of met

19、hods, which in elude the so-called similar day approach,various regressi on models, timeseries, n eural n etworks, statistical lear ning algorithms, fuzzy logic, and expert systems, have bee ndeveloped for short-term forecasti ng.As we see, a large variety of mathematical methods and ideas have bee

20、n used for load forecasti ng. Thedevelopme nt and improveme nts of appropriate mathematical tools will lead to the developme nt of moreaccurate load forecasti ng tech niq ues. The accuracy of load forecast ing depe nds not only on the loadforecasti ng tech niq ues, but also on the accuracy of foreca

21、sted weather sce narios. Weather forecasting is an importa nt topic which is outside of the scope of this chapter.We simply men ti on sig nifica nt progress in the developme nt of computerized weather forecast ingsystems, in clud ing the Mesoscale Model MM5 developed and supported by a con sortium o

22、f universities.2. Importa nt Factors for ForecastsFor short-term load forecast ing several factors should be con sidered,such as time factors, weather data,and possible customers classes. ThBimeriigrtermforecasts take into acco unt the historical load and weather data, the nu mber of customers in di

23、ffere ntcategories, the applia nces in the area and their characteristics in clud ing age, the econo mic anddemographic data and their forecasts, the applia nee sales data, and other factors.The time factorsinclude the time of the year, the day of the week,and the hour of the day. There are importa

24、nt differe ncesin load betwee n weekdays and weeke nds. The load on differe nt weekdays also can behave differe ntly.For example, Mon days and Fridays being adjace nt to weeke nds, may have structurally differe nt loadstha n Tuesday through Thursday. This is particularly true during the summer time.

25、 Holidays are moredifficult to forecast tha n non-holidays because of their relative in freque nt occurre nee. Weather con ditions in flue nee the load. In fact, forecasted weather parameters are the most importa nt factors inshort-term load forecasts.Various weather variables could be con sidered f

26、or load forecasti ng.Temperature and humidity are the most com mon ly used load predictors. Anelectric load predictio n survey published in in dicated that of the 22research reports considered, 13 madeuse of temperature only, 3 made use of temperature and humidity, 3 utilized additi onal weatherpara

27、meters,a nd 3 used only load parameters.Am ong the weather variables listed above, two compositeweather variable functions, the THI (temperaturehumidity in dex) an dWCI (wind chill in dex), are broadlyused by utility compa ni es. THI is a measure of summer heat discomfort and similarly WCI is cold s

28、tressin win ter.Most electric utilities serve customers of differe nt types such as reside ntial,commercial, and in dustrial.The electric usage patter n is differe nt for customers that bel ong to differe nt classes but is somewhatalike for customers within each class. Therefore, most utilities dist

29、inguish load behavior on aclass-by-class basis.3. Forecast ing MethodsOver the last few decades a nu mber of forecast ing methods have bee n developed. Two of the methods,so-called en d-use and econo metric approach are broadly used for medium- and Ion g-term forecast ing.A variety of methods, which

30、 in clude the so-called similar day approach, various regressi on models, timeseries, n eural n etworks, expert systems,fuzzy logic, and statistical learning algorithms, are used forshort-term forecasti ng. The developme nt, improveme nts, and inv estigatio n of the appropriatemathematical tools wil

31、l lead to the developme nt of moreaccurate load forecasti ng tech niq ues.Statisticalapproaches usually require a mathematical model that represe ntsload as fun ctio n of differe nt factorssuch as time, weather, and customer class. The two importa nt categories of such mathematical modelsare: additi

32、ve models and multiplicative models. They differ in whether the forecast load is the sum(additive) of a nu mber of comp onents or the product (multiplicative) of a nu mber of factors.For example,Che n et al.prese nted an additive model that takes the form of predict ing load as the function of fourc

33、omp onen ts:L = Ln + Lw + Ls + Lr, where L is the total load, Ln represents the“ normal ” part of the load,whichis a set ofstandardized load shapes for each“ type ” of day that has been identified as occurringthroughout the year, Lw represe nts the weather sen sitive part of the load, Ls is a specia

34、l eve nt component that create a substa ntial deviati on from the usual load pattern, and Lr is a completely random term,the noise.Chen et al. also suggested electricity pricing as an additional term that can be included in themodel. Naturally, price decreases/increases affect electricity con sumpti

35、 on. Large cost sen sitive industrial and in stituti on al loads can have a sig nifica nt effect on loads. The study in 4 used Penn sylvani a-New Jersey- Maryla nd (PJM) spot price data (as it related to On tario Hydro load) as a n eural network in put. The authors report that accurate estimates wer

36、e achieved more quickly with the inclusionof price data.A multiplicative model may be of the formL = Ln Fw Fs Fr, where Ln is the no rmal (base) load and the correct on factors Fw, Fs, and Fr are positivenu mbers that can in crease or decrease the overall load.These correcti ons are based on curre n

37、t weather (Fw), special eve nts (Fs),a nd ran dom fluctuation (Fr).Factors such as electricity pricing (Fp) and load growth (Fg) can also be in cluded. Rahma n prese nted arule based forecast using a multiplicative model. Weather variables and the base load associated withthe weather measures were i

38、n cluded in the model.3.1 Medium- and Iong-term load forecasting methodsThe en d-use modeli ng, econo metric modeli ng, and their comb in atio ns are the most ofte n usedmethods for medium- and Ion g-term load forecast in g.Descripti ons of applia nces used by customers,the sizes of the houses, the

39、age of equipme nt, tech no logy cha nges, customer behavior,and populationdynamics are usually included in the statistical and simulation models based on the so-called en d-useapproach. In additi on,econo mic factors such as per capita in comes, employme nt levels, an delectricityprices are in clude

40、d in econo metric models. These models areofte n used in comb in ati on with the end-use approach. Lon g-term forecasts in clude the forecasts on the populati on cha nges, econo micdevelopme nt, in dustrial con struct ion, and tech no logy developme nt.En d-use models. The en d-use approach directly

41、 estimates en ergy con sumpti on by using exte nsive informati on on end use and end users, such as applia nces, the customer use, their age, sizes of houses,and so on.Statistical information about customers along with dyn amics of cha nge is the basis for theforecast.End-use models focus on the var

42、ious uses of electricity in the reside ntial,commercial, and industrial sector. These models are based on the prin ciple that electricity dema nd is derived fromcustomer s dema nd for light, cooli ng,heat in g, refrigerati on, etc. Thus en d-use models expla in en ergy dema nd as a fun cti on of the

43、 nu mberof applia nces in the market.Ideally this approach is very accurate. However, it is sensitive to the amountand quality of end-use data. For example, in this method the distributi on of equipme nt age is importa ntfor particular types of applia nces. En d-use forecast requires less historical

44、 data but more in formati onabout customers and their equipme nt.Econo metric models. The econo metric approach comb ines econo mictheory and statistical tech niq uesfor forecasti ng electricity dema nd. The approach estimates the relatio nshipsbetwee n en ergy con sumpti on (depe ndent variables) a

45、nd factors in flue ncing con sumpti on. The relation ships are estimated by the least-squares method or time series methods.One of the options in thisframework is to aggregate the econo metric approach, whe n con sumpti on in differentsectors(residential, commercial, industrial, etc.) is calculated

46、as a function of weather, econo mic andother variables, and the n estimates are assembled using rece nt historical data. In tegrati on of the econometric approach into the en d-use approach in troduces behavioral comp onents into the en d-use equations.Statistical model-based learni ng. The en d-use

47、 and econo metric methods require a large amount of informati on releva nt to applia nces,customers, econo mics, etc. Their application is complicated andrequires human participation. In addition such information is often not available regarding particularcustomers and a utility keeps and supports a

48、 profile of an “ average ” customer or average customersfor different type of customers. The problem arises if the utility wants to con duct n ext-year forecasts forsub-areas, which are ofte n called load pockets. In this case, the amount of the work that should beperformed in creases proporti on al

49、ly with the nu mber of load pockets. In additi on, en d-use profiles andecono metric data for differe nt load pockets are typically differe nt. The characteristics for particularareas may be differe nt from the average characteristics for the utility and may not be available .In orderto simplify the

50、 medium-term forecasts, make them more accurate,a nd avoid the use of the un available informati on,Fein berg et al. developed a statistical model that lear ns the load model parameters from the historicaldata. Fein berg et al. studied load data sets provided by a utility compa ny inNortheaster n US

51、. The focus of the study was the summer data. We compared several load models andcame to the conclusion that the following multiplicative model is the most accurate L(t) = F(d(t), h(t) f(w(t)+ R(t), where L(t) is the actual load at time, d(t) is the day of the week, h(t) is the hour of the day, F(d,

52、 h)is the daily and hourly comp onent, w(t) is the weather data that in clude the temperature and humidity,f(w) is the weather factor, and R(t) is a ran dom error .In fact, w(t) is a vector that con sists of the curre ntand lagged weather variables. This reflects the fact that electric load depe nds

53、 not only on the curre ntweather con diti ons but also on the weather duri ng theprevious hours and days. In particular, the well-k nown effect of the so-called heat waves is that the useof air con diti oners in creases whe n the hot weather continues for several days.To estimate the weatherfactor f

54、(w), we used the regressi on model f(w) =B0 +_BjXj ,where Xj are expla natory variables which are non li near functions of curre nt and past weatherparameters andB0,Begrsstherr coef-ficients.The parameters of the model can be calculated iteratively.We start with F = 1. Then we use the above regressi

55、 on model to estimate f.Then we estimate F, and so on. The described algorithm dem on strated rapid conv erge nee on historicalhourly load and weather data. We have applied it to many areas with population between 50,000 and250,000 customers. Figure 12.1 presents an example of a scatter plot that co

56、mpares the model and realparameters. Figure 12.2 dem on strates the conv erge nee of the correlati onbetween the actual load and the model for the iteration process. Figure 12.3 dem on strates theconv erge nee of the lin ear regressi on procedures in the algorithm.Figure 12.1. Scatter plot of the ac

57、tual load vs the model.The software , that uses the described method, lear ns the model parameters and makes next-yearpredictions based on the model loads for the last 25-30 years of data. Thoughhistorical loads may not available,the software applies the last year models to the historical weather da

58、tato estimate the n ext year s peak distributi on.Figure 122 Correlati on betwee n the actual load and the model.Figure 12.3. Conv erge nee of the R2 for the actual load vs the model.The software gen erates several importa nt characteristics. For example,for each load pocket and for thesystem, it ca

59、lculates a weather normalization factor that is a ratio of the peak load to the load that wouldbe observed un der average peak con diti on s. It also produces probability distributi ons for the n ext yearpeaks.The described methods can be applied to both medium- and Ion gterm forecast ing. However,

60、theIon g-term forecasts should in corporate econo mic and populati on dyn amic forecasts as in putparameters.3.2 Short-term load forecasti ng methodsA large variety of statistical and artificial intelligence techniques have been developed for short-term loadforecasti ng.Similar-day approach. This approach i