Demand Forecasting

CHAPTER-1

EXECUTIVE SUMMARY

1

1.1 Introduction

Power is today a basic human need. It is the critical infrastructure on which

modern economic activity is fully dependent. The power sector is going through a

process of restructuring and reforms in many parts of the world. The driving forces

behind these reforms vary from country to country but the most common are lack

of public resources to finance the required investment for development,

managerial insufficiency, regulatory failures, etc. The political climate and

government policy also shapes the reforms. As a result, the reform models vary

from one another in respect of ownership and management, level of integration,

degree of monopoly and competition, etc. though none is unique. Evolving a

reform program is therefore a State/Country specific task keeping in view the

specific characteristics and developmental needs of its power supply industry as

well as the policies of the government.

In India, such an exercise has already been done in case of Orissa, Delhi,

and the State Electricity Board (SEB) has been restructured. The most

comprehensive piece of legislation, overriding all the existing ones in the power

sector, was passed by both houses of Parliament in April 2003 and enacted in

June 2003. The Electricity Act 2003 provides for increased competition in the

sector by facilitating open access (permission to use the existing power transfer

facilities) to transmission and distribution, power trading and also allowing setting

up of captive power plants without any restriction. The impact of this new Act will

be far-reaching and fundamental. This act will herald new era of the Indian power

sector.

2

Changes in the electricity act and trends toward competition within and between

industries will affect consumers, regulators and utility service providers alike. The

trend expected in the electricity act 2003 is away from public utilities as regulated

monopolies and toward a system in which there is more than one company

offering services to a geographic area. The introduction of new competitive

marketing arrangements aims to:

Extend customer choice and encourage efficiency in the use of electricity;

Encourage efficient investment and innovation in the provision of electricity;

Promote demand management to achieve a supply/demand balance.

Essential to these aims is the development of an accurate real-time

forecasting program. The required system implementation should facilitate use at

various levels of sophistication and operator knowledge and provide accurate

measures of confidence in the forecasts in order that decision making can

proceed.  As the electricity sector is charging up, there is an urgent need for

precision in the demand forecasts. In the past, the world over, an underestimate

was usually attended to by setting up turbine generator plants fired by cheap oil or

gas, since they could be set up in a short period of time with relatively small

investment. On the other hand, overestimates were corrected by variable demand

growth. The underlying notion here was that in the worst case, there would be an

excess capacity, which would be absorbed soon. In the Indian context, the

demands were usually under- estimated, notwithstanding which, the capacities fell

short of the actual demands on a year-to-year basis.

The nature of the forecasts has also changed over the years. It is not

enough to just predict the peak demand and the total energy use, say on an

annual basis. Since the whole objective of demand-side management is to alter

3

the system demand shape, a demand shape forecasting capability within the

system is most desirable. The various methods of implementing demand-side

management are time of use pricing, use of curtail able/interruptible rates,

imposition of penalties for usage beyond a predetermined level, and real time

pricing. A time-of-day tariff structure to manage peaks and troughs in electricity

demand, an hour-by-hour demand shape forecast has become an essential

prerequisite.

In this context short-term demand forecasting also plays a role in the process of

regulation. A precise estimate of demand is important for capacity augmentation,

for accommodating the future demand growth & also helps to schedule the

network management. This report is an effort to comprehend the effect of volatile

demand in the forecasting methods so as to arrive at more accurate estimation of

short-term demand at B.S.E.S networks.

4

1.2 Importance of the Problem:

All though virtually all business functions needs accurate forecast, little

attention is devoted to examine how the forecasting function should be managed.

Competitive markets are defined by supply and demand, so demand is not only

important; it is also particularly tricky when it comes to power. Since power can’t

be stored, plants have to be on when it is needed. And there is only so much

power in a given region, an amount that can be reduced by forced outages and

plant maintenance. It turns out that certain rules of thumb become second nature

to most successful power traders. Industrial demand is relatively constant during

the week. At a times, there are large variations in demand may be due to sudden

shifts in temperature, especially during the summer, when the de facto solution is

to turn on air conditioners. So traders are apt to hover over weather forecasts,

reasoning that there is some strong correlation between shifts in temperature and

shifts in demand. This kind of logic only goes so far. There are problems. The first

is the accuracy of the weather forecast. The second problem concerns some

reliable estimation of demand based on the temperature and other variables. The

third is how demand intersects supply, as the supply curve is discontinuous and

linked to demand in the area. The fourth problem is how power will flow over

transmission lines. Getting it is quite possible to generate a reasonable demand

forecast with a decent temperature forecast and standard statistical techniques.

For long term forecasting, there is a more difficult problem historical

demands are made available on an annual basis. So in those regions, more

5

approximate methods can get utilities with distorted forecasts The other factors

which play important roles are seasonal patterns, similar days in the past,

population changes, and so on.

In India, the power sector has a monopolistic business conditions and thus

lacking for the adoption of the techno-commercial and efficient business practices.

The profit, reliability of supply and the customer satisfaction were not the

buzzwords and thus planning in each area was least considered. These all facts

led to piling up of losses, unplanned blackouts, frequent grid failures, production

loss and ultimately the negative impact on national income. TATA, BSES being

private players in the sector had been profit driven and adopted a professional

management practices in revenue management, network management and other

operational areas. Still by virtue of their monopolistic market position, the pace

adopted was slower as compared to the international standards. Thus the demand

forecasts were the derivative of the traditional methods available and were

sufficient enough to take care of long-term demand. The blackouts and network

failure were common phenomenon observed by the consumers due to poor

demand side management. The short-term and medium term demand forecasting

was no more a choice of these utilities as they were happy with their performance

as compared to SEBs. The Electricity Act 2003 has spurred the overall scenario in

the sector and called for the quality reliable supply to the consumers under section

57 – Standard of Performance in which penalty for the poor supply is imposed. So

it becomes imperative for these private players to focus on demand side

management, which can be better achieved by building up the short-term and

medium term demand forecasting models to be integrated with existing long-term

forecasting techniques.

6

1.3 Scope of the Project

The objective of the project is to

1. Identifying variables for short-term demand forecasting.

2. Developing short-term & peak demand forecasting model.

3. Devising operating procedures for short term forecasting.

4. Suggesting the medium term forecasting model to be operated with the

inputs of short term forecasting model.

Considering the time frame & resources required, the pilot model short-term

demand forecasting model is devised as applicable to the distribution of electricity

in suburbs of Mumbai only.

7

1.4 Outline Of The Project

This project is structured as follows.

Chapter 2 Outlines the brief introduction of B.S.E.S Ltd

Chapter 3 Provides the briefings on the electricity distribution sector in

India & problems faced in demand forecasting.

Chapter 4 Gives insight into literature on demand forecasting. It

discusses the types & methods of demand forecasting & the

methods used in Indian power sector

Chapter 5 Outlines methods of data collection & analysis

Chapter 6 Describes the detached regression analysis for the short-term

demand forecasting model & the pre-dispatch & proposed

Medium term forecasting process

Chapter 7 Discusses the results of study

Chapter 8 Outlines the scope for future work

8

CHAPTER-2

BRIEF PROFILE OF B.S.E.S

9

2.1 Introduction

BSES Limited was incorporated on 1st October 1929, for the distribution of

electricity in the suburbs of Mumbai. Over the years, BSES has grown up to

become India’s premier utility engaged in the generation, transmission and

distribution of electricity. In Mumbai alone, BSES caters to the needs of 2.23

million consumers over an area of 384 sq. km. with a maximum system demand of

approximately 1226 MVA. With 7 decades in the field of power distribution, the

Electricity Supply Division of BSES has achieved the distinction of operating its

distribution network with 99.98% on-line reliability. BSES was amongst the first

utilities in India to adopt computerization in 1967 to meet the increasing

workdemand and to improve services to its customers. With a view to gainfully

utilizing its trained manpower and expertise in the field of power distribution, the

company commenced contracting activities in 1966 by undertaking turnkey

electrical contracts, thermal, hydro and gas turbine installations and

commissioning contracts, transmission line projects etc.

Currently, it serves an area of about 384 sq. km in the suburbs of Mumbai

supplying power to over 2.23 million consumers with a maximum system demand

10

of about 1226 MVA and is the sole distributor of electricity up to 1000 KVA in the

suburbs of Mumbai. As a distribution licensee, the Company has 2829 km of HT

and 2965 km of LT mains on the Company’s System, 51 nos 11 KV distribution-

receiving stations and over 3653 distribution sub-stations. Over 1000 qualified

engineers specialize in installation, testing, operation and maintenance of the

system. BSES enjoys a reputation of being one of the most efficient organization

in the field of electrical transmission and distribution in the country with system

reliability of 99.98 % and T&D losses of about 13% - 14%.

BSES started to set up its own 500 MW coal fired thermal Power Plant and

the first 2 x 250 MW units of Dahanu Power Station were synchronized and began

commercial operation during 1995-1996. A dedicated 220 kV double circuit

transmission line network with three 220/33 kV transmission receiving stations at

Versova, Aarey and Ghodbunder have been installed to transmit the power to the

distribution network area of the Company. During 2002-2003, the station operated

at plant demand factor of 90.53% and achieved the plant availability of over 91%.

The success at Dahanu coupled with long construction, erection and

commissioning experience of BSES has been furthered strengthened by

undertaking successful commissioning of:

1. 220 MW natural gas/ naptha based combined cycle power plant at

Samalkot near Kakinada, Andhra Pradesh in collaboration with APSEB.

2. 165 MW Liquid fuel based Combined cycle power plant at Kochi in Kerala

with Aero-derivative Gas turbines of 43.5 MW size of GE's LM6000 module.

3. 7.59 MW Wind Farm Project comprising of 33 windmills in Chitradurga

District of Karnataka

11

A summary on operational excellence has been given in figure 1 & 2

2.2 Affiliate Companies, Associate & Joint Ventures

A) UTILITY POWERTECH LIMITED

Utility Powertech Limited (UPL), a joint venture with National Thermal Power

Corporation Limited is engaged in undertaking construction, erection, renovation

and modernization and other project management activities in the power sector.

UPL has secured several prestigious jobs in the power and infrastructure sectors

including for renovation and modernization of Boiler of a Thermal Power Station in

Bangladesh.

B) NESCONorth Eastern Electricity Supply Company of Orissa Limited (NESCO) having it’s

head quarters at Balasore, caters to a consumer base of about 0.40 mln covering

an estimated population of 9.18 million on a licensed area of 28,000 sq kms.

C) SOUTHCOSouthern Electricity Supply Company of Orissa Limited (SOUTHCO), having it’s

head quarter at Behrampur, caters to a consumer base of about 0.43 mln covering

an estimated population of 8.71 million in a licensed area of 47,000 sq kms.

12

D) WESCOWestern Electricity Supply Company of Orissa Limited (WESCO) having it’s head

quarter at Burla, caters to a consumer base of about 0.41 million covering an

estimated population of 9.40 million on a licensed area of 48,000 sq kms.

E) ST- BSES COAL WASHERIES LTD.

ST-BSES Coal Washeries Limited is a joint venture of BSES with Spectrum

Technologies and CLI Corporation of USA. It has set up facilities for washing 2.5

million tons of coal per annum at its Coal Washeries at Korba (M.P). The washery

is in operation for almost 2-3 years now and supplies washed coal to Dahanu

Power station, GEB, etc. This coal washery operation helps to remove the ash

content s & thus improve the calorific value of the coal and yields a better power

generation. Also it results into a cost reduction by virtue of reduction in

transportation cost.

F) BSES ANDHRA POWER LIMITEDBSES Andhra Power Limited (BAPL), has set up a 220 MW Dual Fuel (Natural

Gas/Naptha) based Combined Cycle Power Station at Samalkot in Andhra

Pradesh which has started generation from February, 2003 and is currently supply

power to AP Transco under a long term PPA.

G) BSES KERALA POWER LIMITED (BKPL)

13

BKPL has set up a 165 MW naphtha based Combined Cycle Power Station at

Kochi in Kerala. The Combined cycle commercial operation started in November

2000. The plant is currently supplying power to the Kerala State Electricity Board.

H) BSES RAJDHANI POWER LIMITEDBSES has it’s existence Delhi by acquiring the stake in Delhi Vidyut Board post

privatization. In Delhi BRPL is engaged in the business of distribution & supply of

electricity in South & West parts of Delhi covering a consumer base of 8.69 lakhs

spread over 11 districts – Nehru Place, R K Puram, Najafgarh, Alaknanda,

Mehrauli,Palam, Nangloi, Nizamuddin, Janakpuri, Vikaspuri & Punjabi Bagh areas.

I) BSES YAMUNA POWER LIMITEDBYPL is engaged in the business of distribution & supply of electricity in Central &

East parts of Delhi covering a consumer base of 8.49 lakhs spread over 11

districts – Yamuna Vihar, Krishna Nagar, Chandni Chowk, Paharganj, Nand Nagri,

Mayur Vihar, Daryaganj, Jhilmil, Laxmi nagar & Shankar road areas.

14

2.4 Business Objectives & Certifications

● Make available reliable, uninterrupted and quality power to its customers.

● Ensure continuous improvement in systems and processes by incorporating

the latest technology and conforming to national and international

standards.

● Continuously monitor the impact of its operations on the environment and

evolve measures to maintain the ecological balance.

● Develop and maintain a highly motivated, trained and courteous workforce

to understand and meet the changing needs of customers.

● Provide reliable, prompt and economical value added services to its

customers.

B.S.E.S limited has acquired IS0- 9000 & ISO-14000 certification for

Generation, transmission and distribution of electricity.

Service in the field of electrical projects and related construction activities.

Computer services in the field of electricity billing and related applications.

15

Finance, HRD, Secretarial and Investor Service Center (ISO 9002

certification in March 2001).

ISO - 14001 certification for environmental management to Dahanu

Thermal Power Station.

FIG –2.1 Brief profile of B.S.E.S

16

Figure-2.2 Performance of B.S.E.S Mumbai distribution (2001-02)

17

CHAPTER - 3

18

NEED FOR SHORT TERM

FORECASTING MODEL

19

3.1 Power Sector scenario in India

After gaining independence from the British in 1947, the government of

India enacted the Electricity Supply Act of 1948. The 1948 Act brought all new

power generation, transmission, and distribution under the responsibility of the

public sector, especially at the state level. As a result, each state and union

territory established State Electricity Boards (SEBs), vertically integrated entities

that were funded by the state governments. By the early 1990s, the SEBs

controlled 70 percent of the generation, most of the transmission lines, and a

majority of the distribution. Ever since India attained independence, development

of the electricity sector has primarily been the responsibility of the government,

with a relatively small contribution from private enterprises, in the form of licensees

like the BSES (Bombay Suburban Electricity Supply Company), the TEC (Tata

Electricity Company), the CESC (Calcutta Electricity Supply Company), and the

AEC (Ahmedabad Electricity Company).

After sustaining a closed economy since independence, India experienced

a balance of payments crisis in the early 1990s. As part of an effort to liberalize

the economy, an amendment was made to the 1948 Electricity Supply Act, called

the Electricity Laws (Amendment) Act of 1991.

Over the past decade, India have begun major structural reforms of its’

electricity markets. These reforms are aimed at breaking up traditional regional

monopolies and replacing them with several generation and distribution utilities

that bid to sell or buy electricity through a wholesale market. While the rules of

how various wholesale markets operate differ, in each case it is hoped that the

20

end result is a decline in the price of electricity to end users and a price that better

reflects the actual costs involved. One purpose of the legislation was to encourage

private investment in power generation through eight “fast track” projects.

Strategies were being evolved and advice sought from experts, to make the

reform process a success story. One such example is that of Arthur Andersen, the

management consultant. According to a paper presented at the FICCI (Federation

of Indian Chambers of Commerce and Industry) seminar on Restructuring of

PSUs, Arthur Andersen pointed out that

● In the first phase, the government should establish an appropriate legal and

regulatory framework and un-bundle the SEBs into generation,

transmission, and distribution, set-ups.

● The second phase of the reform, the report adds, should focus on

corporatisation, commercialization and institutional strengthening of the

entities, privatization of part of the distribution entities, partial restoration of

credit worthiness of the sector and improvement in system efficiency.

● Thereafter, the study suggested that the government should complete

privatization of distribution business. This has to be accompanied by

introduction of private investment to generation through competitive

solicitation for individual power producers and consolidation of the

functioning and financial performance of the new power utilities.

● In the last phase, the report stressed that efforts should be made to achieve

higher customer satisfaction through reduction of power deficit, attainment

21

of higher quality and efficiency in electricity services through increased

competition and private participation.

3.1.1. The electricity Act-2003 & it’s impact on B.S.E.S

Then came the most comprehensive piece of legislation, overriding all the

existing ones in the power sector, was passed by both houses of Parliament in

April 2003 and enacted in June 2003. Electricity Act provides for increased

competition in the sector by facilitating open access (permission to use the

existing power transfer facilities) to transmission and distribution, power trading

and also allowing setting up of captive power plants without any restriction. The

states are required to align their reform. The impact of this new Act will be far-

reaching and fundamental.

The latest bill, the electricity bill 2003 seeks to bring about a qualitative

transformation of the electricity sector through a new paradigm. The Bill seeks to

create liberal framework of development for the power sector by distancing

Government from regulation. Looking at the act the power sector will have entry of

more players (mostly private, some public) into generation and distribution and will

become more complex with entry of many more actors and contracts. Group

captive, private distribution companies, transmission licensees and power traders

are some new actors. With open access, TOD tariff, many supplier- trader -

consumer contracts and many dispersed systems, planning, regulating and

operation of the system becomes more complex.

The availability-based tariff (ABT) has compelled the utilities to devise their

medium & short-term demand and avoid the grid failure incidents by drawing extra

needed power at peak hours. The heavy penalty built in the ABT, intends to

22

govern the grid discipline and ensure the reliability of the system. The utilities are

therefore forced to forecast their demand on weekly, daily and, probably in near

future, hourly basis to avoid the penalty.

A detailed consumer category-wise consumption, short term forecast helps

in the determination of a just and reasonable tariff structure wherein no consumer

pays less than the cost incurred by the utility for supplying the power. Also, the

utility can then plan the power purchase requirements so as to meet the demand

while maintaining the merit order dispatch to achieve optimization in the use of

their resources. The presence of economies of scale, lesser focus on

environmental concerns, predictability of regulation and a favorable public image,

all made the process of forecasting demand much simpler. In contrast, today an

underestimate could lead to under capacity, which would result in poor quality of

service including localized brownouts, or even blackouts. An overestimate could

lead to the authorization of a plant that may not be needed for several years.

Moreover, in view of the ongoing reform process, with associated unbundling of

electricity supply services, tariff reforms and rising role of the private sector, a

realistic assessment of demand assumes ever-greater importance. These are

required not merely for ensuring optimal phasing of investments, a long-term

consideration, but also rationalizing pricing structures and designing demand side

management programs, which are in the nature of short- or medium-term needs.

So at this juncture it became essential for B.S.E.S to select a short term

demand forecasting model having focused on customer & electricity of good

quality be provided at reasonable rates for economic activity so that

competitiveness increases. This demands for the reliability of supply at the

reasonable price and thus necessitates a planning of demand and supply

23

positions so as to zeroing the blackouts. A suitable forecasting model for short

term and medium term demands has thus become a need of the hour as the long

term forecasting traditional models of are insufficient to comprehend the erratic

changes in the consumer demand.

By virtue of the norms specified in Standard of Performance (SOP) for the service

quality, the medium & short term - demand forecasts will be significant and if not

adhered to will lead to the revenue erosion by way of penalty payable to statutory

regulating authorities and consumers.

24

3.2 The electricity demand forecasting problem

To successfully operate in these new markets electricity utilities face two complex

statistical problems

● How to forecast both electricity demand and

● The wholesale spot price of electricity.

Failure to implement efficient solutions to these two forecasting problems can

directly result in losses through uninformed trades in the wholesale market. So it

become imperative for the utility to understand the factors affecting demand before

adopting a framework for a demand forecasting model

3.2.1 Factors affecting Demand

Generally, the demand of an electric utility is composed of very different

consumption units. A large part of the electricity is consumed by industrial

activities. Residential consumers in forms of heating, lighting, cooking, laundry, etc

use another part. Also many services offered by society demand electricity, for

example street lighting, railway traffic etc.

1. Factors affecting the demand depend on the particular consumption unit.

The level of the production mostly determines the industrial demand.

The demand is often quite steady, and it is possible to estimate its

dependency on different production levels. However, from the point of view

of the utility selling electricity, the industrial units usually add uncertainty in

the forecasts. The problem is the possibility of unexpected events, like

25

machine breakdowns or strikes, which can cause large unpredictable

disturbances in the demand level.

2. In the case of private people, the factors determining the demand are much

more difficult to define. Each person behaves in his own individual way, and

human psychology is involved in each consumption decision. Many social

and behavioral factors can be found. For example, big events, holidays,

even TV-programs, affect the demand.

3. The weather is the most important individual factor, the reason largely

being the electric heating of houses, which becomes more intensive as the

temperature drops. As a large part of the consumption is due to private

people and other small electricity customers, the usual approach in demand

forecasting is to concentrate on the aggregate demand of the whole utility.

4. In the short run, the meteorological conditions cause large variation in this

aggregated demand. In addition to the temperature, also wind speed, cloud

cover, and humidity have an influence. In the long run, the economic and

demographic factors play the most important role in determining the

evolution of the electricity demand.

5. From the point of view of forecasting, the time factors are essential. Various

seasonal effects and cyclical behaviors (daily and weekly) as well as

occurrences of legal and religious holidays play a key role in forecasting.

6. The other factors causing disturbances can be classified as random

factors. These are usually small in the case of individual consumers,

although large social events and popular TV-programs add uncertainty in

the forecasts. Industrial units, on the other hand, can cause relatively large

disturbances.

26

3.2.2: Issue of data

Since power can’t be stored, plants have to be on when it is needed. And

there is only so much power in a given region, an amount that can be reduced by

forced outages and plant maintenance. It turns out that certain rules of thumb

become second nature to most successful power traders. Industrial demand is

relatively constant during the week. What cause large variations in demand are

sudden shifts in temperature, especially during the summer when the de facto

solution is to turn on air-conditioners. So utilities are apt to hover over weather

forecasts, reasoning that there is strong correlation between shifts in temperature

and shifts in price. This kind of logic only goes so far. There are problems.

1. The first is the accuracy of the weather forecast.

2. The second problem concerns some reliable estimation of demand based

on the temperature and other variables.

3. The third is how demand intersects supply, as the supply curve is

discontinuous and linked to fuel types and prices

4. The fourth problem is how power will flow over transmission lines.

Getting a complete picture of power fundamentals is just plain difficult.

While there are solutions to getting at least parts of this picture, the point here is to

address the second problem. The problem with predicting demand is the scarcity

of data, because the only way to get a truly accurate demand forecast is to know

the near-real-time demand in a region. Many times historical demands are made

available on an annual basis. So in such cases, more approximate methods can

get the utility into a bigger ballpark, with variation caused by current temperature

forecasts. Lots of considerations go into this, including seasonal patterns, similar

days in the past, population changes, and so on.

27

Exotic techniques for predicting demand often appear in technical literature, citing

neural nets or other methods. A neural net is a computer forecasting technique

based on learning through patterns of behavior. Neural nets are known for being

insensitive to rapid changes, but besides that, most of these techniques have a

high sensitivity to available data. The second issue is that accuracy of the forecast

increases with the size of the database used to arrive at the model. However,

there is a practical limit to the quantity of data that is cost effective to gather and

manipulate in terms of additional information gained by the utility planner. Studies

are on, the world over, to develop new techniques that would reduce the amount

of data required for a given level of accuracy to be achieved in the forecast.

28

3.3. Conclusion

The Electricity Act 2003 has spurred the momentum in the power sector by

encompassing all the aspect with an impact on utility’s balance sheet, so as to take

a cognizance of the service reliability as desired by the consumer. This may not be

a panacea for the problems of power sector as the exercise may encounter to

different issues as discussed, but will definitely imbibe innovative changes in the

existing processes. This will necessitate the better demand side management, and

short term & medium term demand forecast will enable to resolve the issues

arising at the execution level.

29

Chapter-4

LITERATURE REVIEW ON

DEMAND FORECASTING

30

4.1 Introduction

Econometric methods of forecasting, in the context of energy demand

forecasting can be described as ‘the science and art of specification,

estimation, testing and evaluation of models of economic processes’ that

drive the demand for fuels. 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

imperative to use modeling techniques which capture the effect of factors such as

prices, income, population, technology and other economic, demographic, policy

and technological variables.

Over the last few decades a number of forecasting methods have been

introduced. Most of these methods use statistical techniques sometimes combined

with artificial intelligence algorithms such as neural networks, fuzzy logic, and

expert systems. Two of the methods, so-called end-use and econometric

approach are broadly used for medium and long-term forecasting. A variety of

methods, that include the so-called similar day approach, various regression

models, time series, neural networks, statistical learning algorithms, fuzzy logic,

and expert systems, have been developed for short-term forecasting.

31

A large variety of mathematical methods and ideas have been used for demand

forecasting. The development and improvements of appropriate mathematical

tools will lead to the development of more accurate demand forecasting

techniques. The accuracy of demand forecasting depends not only on the demand

forecasting techniques but it also depends on the accuracy of forecasted weather

scenarios.

4.1.1 Types of demand forecasting

Based on the forecasting period, demand forecasting can be categorized

into Long Term Demand (LTL) Forecasting (up to 20-year period), Medium Term

Demand (MTL) forecasting (2-year period), STL forecasting (30-minute period)

and VSTL forecasting (5-minute period). Each of these classes focuses on

different criteria to consider.

1. STL & VSTL: which are usually from one hour to one week, medium

forecasts which are usually from a week to a year, and long-term forecasts

which are longer than a year. The forecasts for different time horizons are

important for different operations within a utility company. The natures of

these forecasts are different as well. For example, for a particular region, it

is possible to predict a next day demand with an accuracy of approximately

1-3%.

2. MTL: However, it is impossible to predict the next year peak demand with

the similar accuracy since accurate long- term forecasts are not available.

For the next year peak forecast, it is possible to provide the probability

distribution of the demand based on historical weather observations. It is

also possible, according to the industry practice, to predict the so-called

32

weather normalized demand, which would take place for average annual

peak weather conditions or worse than average peak weather conditions for

a given area.

3. LTL: Demand forecasting is also important for contract evaluations and

evaluations of various sophisticated financial products on energy pricing

offered by the market. In the deregulated economy, decisions on capital

expenditures based on long-term forecasting are also more important than

in a non-deregulated economy when rate increases could be justified by

capital expenditure projects.

Short-range planning is motivated by a need to reach a decision, to commit

to a particular installation or type of construction, and to do so at the lead-time.

Letting the lead-time for additions in system network without making appropriate

plans is perhaps the worst type of planning mistake. Therefore, what is needed

most in short-range planning is a reliable “alarm” about when present facilities will

become insufficient.

Long-range planning needs are somewhat different. No commitment need

to be made to the elements in a long-range plan, so timing is less important than

in short-range planning. Since the long-range plan evaluates whether and how

short-range commitments fit well into long-range needs, capacity and location are

more important than timing. Unlike short-range planning, long-range T&D planning

requirements for a spatial demand forecast are oriented less toward “when” and

more toward “how much” For long-range planning, knowing what will eventually be

needed is more important than knowing exactly when.

33

4.2 Existing methods

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 of the forecasts. An approach

often used is to employ more than one method and then to compare the forecasts

to arrive at a more accurate forecast. Two of the methods, so-called end-use and

econometric approach are broadly used for medium and long-term forecasting.

A variety of methods, that include the so-called similar day approach, various

regression models, time series, neural networks, expert systems, fuzzy logic, and

statistical learning algorithms, are used for short-term forecasting. The

development, improvements, and rigorous analysis of appropriate mathematical

tools will lead to the development of more accurate demand forecasting

techniques.

Statistical approaches usually require a mathematical model that models

demand as function of different factors such as time, weather, and customer class.

The two important categories of such mathematical models are: additive models,

and multiplicative models. They differ in whether the forecast demand is the sum

(additive) of a number of components or the product (multiplicative) of a number of

factors.

For example, following model presented an additive model that takes the form of

predicting demand as the addition of four components:

L = Ln + Lw + Ls + Lr

Where L is the total demand,

34

Ln - represents the "normal" part of the demand, which is a set of

standardized demand shapes for each "type" of day that has been

identified as occurring throughout the year,

Lw -represents the weather sensitive part of the demand,

Ls - is a special event component that create a substantial deviation from

the usual demand pattern,

Lr - is a completely random term, the noise.

Another variable term that might be included in the model is electricity

pricing. Naturally, price decreases/increases affect electricity consumption. Large

cost sensitive industrial and institutional demands can have a significant effect on

their consumption patterns.

A multiplicative model may be of the form

L = Ln ¢ Fw ¢ Fs ¢ Fr;

Where Ln is the normal (base) demand

FW - is the correction factors,

Fs and Fr is positive numbers that can increase or decrease the overall

demand. These corrections are based on current weather (Fw), special

events (Fs), and random fluctuation (Fr).

4.2.1 Short-term demand forecasting methods

A large variety of statistical and artificial intelligence techniques have been

developed for short-term demand forecasting which are described in following

pages.

35

4.2.1.1-Similar Day Approach: This approach is based on searching historical

data for days within one, two, or three years with similar characteristics to the

forecast day. Similar characteristics include weather, day of the week, and the

date. The demand of a similar day is considered as a forecast. Instead a single

similar day demand, the forecast can be a linear combination or regression

procedure that can include similar several days. The trend coefficients can be

used for similar days in the previous years.

4.2.1.2-Regression Methods: Regression is the one of most widely used

statistical techniques. For electric demand forecasting regression methods are

usually used to model the relationship of demand consumption and other factors

such as weather, day type, and customer class. The models can be incorporated

with deterministic influences such as holidays, stochastic influences such as

average demands, and exogenous influences such as weather.

4.2.1.3-Time Series: Time series methods are based on the assumption that the

data have an internal structure, such as autocorrelation, trend or seasonal

variation. Time series forecasting methods detect and explore such a structure.

Time series have been used for decades in such fields as economics, digital

signal processing as well as electric demand forecasting. In particular, ARMA

(autoregressive moving average), ARIMA (autoregressive integrated moving

average), ARMAX (autoregressive moving average with exogenous variables),

and ARIMAX (autoregressive integrated moving average with exogenous

variables) are the most often used classical time series methods. ARM models are

usually used for stationary processes while ARIMA is an extension of ARMA to no

36

stationary processes. ARMA and ARIMA use the time and demand as the only

input parameters. Since demand generally depends on the weather and time of

the day, ARIMAX is the most natural tool for demand forecasting among the

classical time series models.

4.2.1.4-Neural Networks. The use of artificial neural networks (ANN or simply

NN) has been a widely studied electric demand forecasting technique since 1990).

Neural networks are essentially non-linear circuits that have the demonstrated

capability to do non-linear curve fitting. The outputs of an artificial neural network

are some linear or non-linear mathematical function of its inputs. The inputs may

be the outputs of other network elements as well as actual network inputs. In

practice network elements are arranged in a relatively small number of connected

layers of elements between network inputs and outputs. Feedback paths are

sometimes used. In applying a neural network to electric demand forecasting, one

must select one of a number of architectures, the number and connectivity of

layers and elements, use of bi-directional or unidirectional links and the number

format (e.g. binary or continuous) to be used by inputs, outputs and internally. The

most popular artificial neural network architecture for electric demand forecasting

is back propagation. Back propagation neural networks use continuously valued

functions and supervised learning. That is, under supervised learning, the actual

numerical weights assigned to element inputs are determined by matching

historical data (such as time and weather) to desired outputs (such as historical

electric demands) in a pre-operational "training session". Artificial neural networks

with unsupervised learning do not require pre-operational training.

Input variables in ANN are historical hourly demand data, temperature, and the

day of the week. Extended ANN a demand forecasting system known as

37

ANNSTLF are also used. ANNSTLF is based on multiple ANN strategy that

captures various trends in the data. In the development they used a multiplayer

perceptron trained with error back propagation algorithm. ANNSTLF can consider

the effect of temperature and relative humidity on the demand. It also contains

forecasters that can generate the hourly temperature and relative humidity

forecasts needed by the system. In the new generation, ANNSTLF includes two

ANN forecasters, one predicts the base demand and the other forecasts the

change in demand. The final forecast is computed by adaptive combination of

these forecasts. The effect of humidity and wind speed are considered through a

linear transformation of temperature.

4.2.1.5-Expert Systems. Rule based forecasting makes use of rules, which are

often heuristic in nature, to do accurate forecasting. Expert systems, incorporates

rules and procedures used by human experts in the field of interest into software

that is then able to automatically make forecasts without human assistance.

Expert system use began in the 1960's for such applications as geological

prospecting and computer design. Expert systems work best when a human

expert is available to work with software developers for a considerable amount of

time in imparting the expert's knowledge to expert system software. Also, an

expert's knowledge must be appropriate for codification into software rules (i.e. the

expert must be able to explain his/her decision process to programmers). An

expert system may codify up to hundreds or thousands of production rules.

38

4.2.1.6-Fuzzy Logic: Fuzzy logic is a generalization of the usual Boolean logic

used for digital circuit design. An input under Boolean logic takes on a truth-value

of "0" or "1" (i.e. 1 False or True). Under fuzzy logic an input has associated with it

a certain qualitative ranges. For instance a transformer demand may be "low",

"medium" and "high". Fuzzy logic allows one to (logically) deduce outputs from

fuzzy inputs. In this sense fuzzy logic is one of a number of techniques for

mapping inputs to outputs (i.e. curve fitting). Among the advantages of the use of

fuzzy logic is the absence of a need for a mathematical model mapping inputs to

outputs and the absence of a need for precise (or even noise free) inputs. With

such generic conditioning rules, properly designed fuzzy logic systems can be

very robust when used for forecasting. Of course in many situations an exact

output (e.g. precise 12PM demand) is needed. After the logical processing of

fuzzy inputs, a "defuzzification" process can be used to produce such precise

outputs.

4.2.1.7-Support Vector Machines: Support Vector Machines are a more recent

powerful technique for solving classification and regression problems. This

approach was originated from Vapnik's statistical learning theory. Unlike neural

networks, which try to define complex functions of the input feature space, support

vector machines perform a nonlinear mapping (by using so called kernel functions)

of the data into a high dimensional (feature) space. Then support vector machines

use simple linear functions to create linear decision boundaries in the new space.

The problem of choosing an architect for a neural network is replaced here by the

problem of choosing a suitable kernel for the support vector machine

39

4.2.2 Demand forecasting methods for LTL & MTL

4.2.2.1-Trend method: This method falls under the category of the non-causal

models of demand forecasting that do not explain how the values of the variable

being projected are determined. Here, we express the variable to be predicted

purely as a function of time, rather than by relating it to other economic,

demographic, policy and technological variables. This function of time is obtained

as the function that best explains the available data, and is observed to be most

suitable for short-term projections.

This method has been used by the 16th Electric Power Survey (EPS) of the

Central Electricity Authority to forecast the consumption of most consumer

categories except HT Industries. The Base Paper of the EPS, detailing the

methodological issues, states that in the domestic, commercial and miscellaneous

categories, the observed time series in the number of consumers and

consumption per capita have been projected into the future, with adjustments for

increase in appliance ownership. It is only for the HT industries that an end-use

method is used. It also mentions that adjustments have been made to account for

unmet demands due to the presence of power cuts, though the specific

assumptions have not been elaborated upon. Thus, unrestricted demands were

worked out for the future. The trend method has the advantage of its simplicity and

ease of use. However, the main disadvantage of this approach lies in the fact that

it ignores possible interaction of the variable under study with other economic

factors.

For example, the role of incomes, prices, population growth and

urbanization, policy changes etc., are all ignored by the method. The underlying

notion of trend analysis is that time is the factor determining the value of the

40

variable under study, or in other words, the pattern of the variable in the past will

continue into the future. Therefore, it does not offer any scope to internalize the

changes in factors such as the effects of government policy (pricing or others),

underlying institutional structure, regulatory regimes, demographic trends,

aggregate and per capita growth in incomes, technological developments etc.

However, this method is important as it provides a preliminary estimate of the

forecasted value of the variable. It may well serve as a useful cross check in the

case of short-term forecasts.

4.2.2.2-End-use method: The end-use approach attempts to capture the impact

of energy usage patterns of various devices and systems. The end-use models for

electricity demand focus on its various uses in the residential, commercial,

agriculture and industrial sectors of the economy. For example, in the residential

sector electricity is used for cooking, air conditioning, refrigeration, lighting, and in

agriculture for lift irrigation. The end-use method is based on the premise that

energy is required for the service that it delivers and not as a final good. The

following relation defines the end use methodology for a sector:

E = S x N x P x H

E = energy consumption of an appliance in kWh

S = penetration level in terms of number of such appliances per customer

N = number of customers

P = power required by the appliance in kW

H = hours of appliance use.

This, when summed over different end-uses in a sector, gives the

aggregate energy demand. This method takes into account improvements in

41

efficiency of energy use, utilization rates, inter-fuel substitution etc., in a sector as

these are captured in the power required by an appliance, P. In the process the

approach implicitly captures the price, income and other economic and policy

effects as well.

For instance, the Planning Commission had used a combination of the trend and

end-use methods to forecast the energy requirement in the various sectors. The

results from trend and regression analysis were compared with those obtained

from the end use model to arrive at the best forecasts. Also, detailed sectoral

analysis was carried out, especially for the transport and domestic sectors. To

estimate the end-use model the Planning Commission used, a spreadsheet based

integrated energy demand model called DEFENDUS (Development of End Use

Energy Scenarios). In the household sector, the estimates for energy requirement

for lighting were derived from the kerosene requirement norms and for cooking

and space heating were derived from the useful energy per person per. The

agricultural demand was assessed on the basis of the stock of tractors and pump

sets used while the commercial demand was assumed to follow on lines similar to

the one followed for the household sector. The end-use approach is most effective

when new technologies and fuels have to be introduced and when there is lack of

adequate time-series data on trends in consumption and other variables.

However, the approach demands a high level of detail on each of the end-uses.

4.2.2.3-Econometric approach: This approach combines economic theory with

statistical methods to produce a system of equations for forecasting energy

demand. Taking time-series or cross-sectional/pooled data, causal relationships

could be established between electricity demand and other economic variables.

42

The dependant variable, in our case, demand for electricity, is expressed as a

function of various economic factors. These variables could be population, income

per capita or value added or output (in industry or commercial sectors), price of

power, price(s) of alternative fuels (that could be used as substitutes), proxies for

penetration of appliances/equipment (capture technology effect in case of

industries) etc. Thus, one would have:

ED = f (Y, Pi, Pj, POP, T) Where,

ED = electricity demand

Y = output or incomes

Pi = own price

Pj = price of related fuels

POP = population

T = technology

Several functional forms and combinations of these and other variables

may have to be tried till the basic assumptions of the model are met and the

relationship is found statistically significant.

The econometric methods require a consistent set of information over a

reasonably long duration. This requirement forms a pre-requisite for establishing

both short-term and long-term relationships between the variables involved. Thus,

for instance, if one were interested in knowing the price elasticity of demand, it is

hard to arrive at any meaningful estimates, given the long period of administered

tariffs and supply bottlenecks. However, the price effect will have an important role

to play in the years to come. In such a case, one may have to broaden the set of

explanatory variables apart from relying on more rigorous econometric techniques

to get around the problem. Another criticism of this method is that during the

43

process of forecasting it is incorrect to assume a particular growth rate for the

explanatory variables. Further, the approach 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 behavior of the variable being explained. This would have

to be built into the model, maybe in the form of structural changes.

4.2.2.4- Time series methods

A time series is defined to be an ordered set of data values of a certain

variable. Time series models are, essentially, econometric models where the only

explanatory variables used are lagged values of the variable to be explained and

predicted. The intuition underlying time-series processes is that the future

behavior of variables is related to its past values, both actual and predicted, with

some adaptation/adjustment built-in to take care of how past realizations deviated

from those expected. Thus, the essential prerequisite for a time series forecasting

technique is data for the last 20 to30 time periods. The difference between

econometric models based on time series data and time series models lies in the

explanatory variables used. It is worthwhile to highlight here that in an econometric

model, the explanatory variables (such as incomes, prices, population etc.) are

used as causal factors while in the case of time series models only lagged

(or previous) values of the same variable are used in the prediction. In general, the

most valuable applications of time series come from developing short-term

forecasts, for example monthly models of demand for three years or less.

Econometric models are usually preferred for long-term forecasts. Another

advantage of time series models is their structural simplicity. They do not require

collection of data on multiple variables. Observations on the variable under study

44

are completely sufficient. A disadvantage of these models, however, is that they

do not describe a cause-and-effect relationship. Thus, a time series does not

provide insights into why changes occurred in the variable.

Often in analysis of time series data, either by using econometric methods

or time series models, there do exist technical problems wherein more than one of

the variables is highly correlated with another (multi-collinearity), or with its own

past values (auto-correlation). This sort of a behavior between variables that are

being used to arrive at any forecasts demands careful treatment prior to any

further analysis. These, along with other similar methodological options, need a

careful assessment while working out forecasts of demand for any sector.

4.2.2.5-Multiple Scenario Forecasting

The chief reason for the multi-forecast approach is that it is often impossible

to predict some of the events that will shape the location of long-range distribution

demand growth. In such cases, the planner is advised to admit that these factors

cannot be forecasted, and to run several “what if” studies, analyzing the

implications of each even as it bears on the distribution plan. This is called multiple

forecasting.

The location aspect of electric demand growth is often very sensitive to

issues that simply cannot be predicted with any dependability. As an example,

consider a new bridge that might be planned to span the lake shown to the

northwest of the city, and the difference it would make in the developmental

patterns of the surrounding region. That lake limits growth of the city toward the

northwest. People have to drive around it to get to the other side and there is a

considerable commuting, making land on the other side unattractive as compared

45

to other areas of the city. Now if the bridge is complete it will change the shape of

the city’s demand growth, opening up new areas to development and drawing

demand growth away from other areas. The impact of the bridge will lead to

1. It increases growth to the northeast,

2. And it decreases growth in the eastern/southern portion of the system.

In such situations the planner may be better off admitting that he cannot

forecast accurately the casual event(s), and do “what if” planning to determine the

possible outcomes of the even. By doing multiple-scenario forecasts that cover the

variety of possible outcomes, the planner can alert himself to the consequences of

such unpredictable events. At the very least, he can then watch these events with

interest, against the day a decision must be made, knowing what impact they will

have and having an idea how he will respond with changes in his plan. Hopefully

though, he can rearrange those elements of his plan that are sensitive to the even,

to minimize his risk, perhaps pushing particularly scenario-sensitive elements into

the future and bringing forward other, less sensitive items.

The challenge facing the planner is to develop a short-range plan that

doesn’t risk too much in the event that one eventuality or the other occurs. He

must develop a short-range plan that can economically “branch” toward either

eventuality – a plan with recourse. Such planning is difficult, but not impossible. a

foundation for analysis of risk among different scenarios.

46

4.3: Forecasting methods & problems in Indian power sector

Awareness regarding the importance of demand forecasting has been

increasing in the very recent past. The Uttar Pradesh Electricity Regulatory

Commission in their publication, Power Diary, has recognized the need for the use

of rigorous methods. The tariff order passed by the Commission also highlights the

need for data collection on various essential variables to enable introduction of

further refinements in the future, thus making the reform process faster and more

effective.

The Haryana Electricity Regulatory Commission has issued guidelines

making it mandatory for the licensee to submit to the Commission a forecast of

future demand in the respective areas of supply of each licensee for a period of 10

years. The licensee is also required to furnish details of the data, and assumptions

on which it is based along with a justification for its choice of forecasting

methodology. The Commission reserves the powers to specify, from time to time,

particular matters that should be dealt with in the demand forecasts. These may

include demand-side management programs and anticipated increase in end-use

efficiency, category-wise demands, losses (technical and non-technical), and

seasonal and time-of-day changes in demand shape, benchmarking demand

growth, and other related issues.

In the case of Andhra Pradesh, the quality of data was an issue taken up by

various people in their objections. Only 41% of the total electricity generated is

metered and billed. This leads to inefficiencies in the usage of electricity in the

state. The uncertainty surrounding the estimate of the agricultural consumption

(being un-metered), which in turn has a bearing on the identification of

transmission and distribution losses was also questioned with objectors producing

47

calculations showing that the demand estimates were either over- or under-

estimates. The licensee here relied on various methods to arrive at estimates of

agricultural consumption over the years. For the year 2000-01, the licensee has

estimated monthly consumption pattern aggregating to 9800 MUs. This was

supported by short term forecast of energy sales, energy requirements and peak

demand which projected to 9420 MUs sales while on the other hand estimating

consumption of 9900 MUs on the basis of the agricultural consumption pattern in

Kuppan Rural Electric Co-operative Society. There was no apparent consistency

in the methods used to project the agricultural consumption. Thus, the

Commission directed the licensee to institute a year long sample study during

2000-01 to determine the energy consumption by the agricultural pump sets

covering all the Mandals and ½% of the pump sets in the State apart from the

study taken up by APTRANSCO with funding by the World Bank. The licensee has

also been directed to carry out a census of agricultural pump sets within six

months of the issue of the order.

In Orissa, the role of regulators has been clearly defined by the Electricity

Regulatory Commission Act, 1998. The Regulatory Commission is envisaged to

plays a key role in tariff filing. The Commission has been empowered under

Section 22 (1)(a), and Section 22(2)(c) to regulate the electricity sector in the

state. Thus, the Commission may advise the licensee to take up demand

forecasting exercises to ensure a just and reasonable tariff structure. Planning

strategies for conservation, demand distribution and implementation of other such

demand side management programs will also be aided by a detailed demand

forecast. Besides, a long-term demand forecast would assist the Commission to

evaluate the need and feasibility of investment in building future capacity, keeping

48

in mind the long gestation periods of plants and rapidly changing technologies in

the power sector.

The nature of the forecasts has also changed over the years. It is not

enough to just predict the peak demand and the total energy use, say on an

annual basis. Since the whole objective of demand-side management is to alter

the system load shape, a load shape forecasting capability within the system is

most desirable. The various methods of implementing demand-side management

are time of use pricing, use of curtail able/interruptible rates, imposition of

penalties for usage beyond a predetermined level, and real time pricing. A time-of-

day tariff structure to manage peaks and troughs in electricity demand, an hour-

by-hour load shape forecast has become an essential prerequisite. Further, the

end-use components of the load shape must also be known in order to plan the

other demand-side management activities to achieve maximum conservation,

while avoiding undue demand restrictions.

Another use for demand forecasting models is the assessment of the

impact that a new technology might have on the energy consumption. This helps

planners to evaluate the cost effectiveness of investing in the new technology and

the strategy for its propagation. The use of a straightforward engineering end-use

approach that focuses only physical factors can miss the emergence of new end

uses, as well as other effects such as the impact of rising energy prices as a

stimulus to energy efficiency. Also the process of projecting the demand would

require estimating market penetration of various devices, while accounting for fuel

substitution, average capacity and efficiency factors in the future, as well as

average utilization rates. The demand forecasts are also done for each consumer

category and voltage level. Charging the commercial, industrial and large

49

consumers a higher charge, which is used to subsidize social reform programs,

optimizes revenues while keeping social objectives in mind. The forecast plays an

important role in identifying the categories, which “can pay”, and those that should

be “subsidized”. To deal with all of the above many forecasting techniques have

been developed, ranging from very simple extrapolation methods to more complex

time series techniques, extensive accounting frameworks and optimization

methods, or even hybrid models that use a combination of these for purposes of

prediction.

Given the changing power scenario in the country, a move to better

forecasting methods cannot be undermined. Indeed, while one would face

problems with availability of information, initiating work with better methods would

provide the impetus for collection and collation of the information that would form

the backbone of any useful exercise. For example, the Indian Market

Demographic Surveys conducted by the National Council of Applied Economic

Research (NCAER) provides a time-series on income distribution of population,

appliance ownership by income class, region, profession and the rate of

depreciation, all of which will be critical in determining residential electricity

demand. This could be broadened in scope. Similarly, the household expenditure

surveys could be adapted to meet the demand of energy demand forecasts.

Specifically, the effective use of end-use methods would require that extensive

primary-level surveys be carried out to build and consolidate a reliable database

on end-uses of energy in the different sectors.

It is also noteworthy that the time series data available in India is admittedly

unreliable. In particular, the data on agricultural consumption, being un-metered, is

recorded as residual consumption. With the deregulation, it is now evident that this

50

category absorbs a degree of the technical and non-technical power losses of the

utility to keep these at acceptable levels. A greater transparency in documentation

of sector-wise consumption is clearly impending, especially to provide a boost to

private sector initiative in generation, transmission and sales of electricity through

distribution.

Another issue that needs to come into the limelight is that of unmet demand

for energy and the gap that exists between demand and supply of power in certain

geographical areas. Studies to collect and assimilate data on the unrestricted

demand for power, giving due recognition to non-utility power generation (e.g.

small-scale captive generation by industries, use of generators for residential and

commercial use) are critical. The data on this is inadequate, to say the least.

It is worth mentioning that all the forecasting methods available today are highly

data intensive.

However, as matters stand, in spite of the several uses and advantages of

good demand forecasting methods that have been studied and applied by experts

the world over, a bulk have yet to gain a level of recognition in India. As is

admitted by many, a Regulatory Commission, the estimates of electricity demand

made by the electricity boards are often found to lack the expected degree of

accuracy and rigor. The current practice to estimate demand is to use

compounded annual growth rates (CAGR) or at best a simple trend analysis.

These methods, however, only forecasts demand met and not actual demand

besides ignoring the effect of changes in incomes, prices, consumer tastes and

quality of supply. There is, nonetheless, concern about the cost and feasibility of

modeling electricity demand, in a purely econometric setting, among the utilities in

India.

51

4.4 Conclusion

The need of the short term –forecasting model has become integral part of

the power sector business, so as to meet the challenges of the changing scenario.

The existing traditional practices focus on long term forecasting and thus have a

macro-coverage at the transmission & generation level. Whereas the day-to day

demand of electricity will be derived from the short term forecasting methods

which necessitates an eye for the details.

The real challenge is to develop a reliable short-term forecasting model in

the face of the issues such as non-reliable data availability, dependence on

whether forecast and other non-controllable variables.

52

CHAPTER- 5

DATA COLLECTION &

ANALYSIS

53

5.1 Introduction

The system load is the sum of all the individual loads. The two essential

components of a system load are:

Base component.

Randomly variable component.

The proposed model of Short-term load forecasting provides data of base

component for each hour and covers a period of one week. The base component

data comprises of

1. Hourly or half-hourly peak load (KW)

2. Hourly or half-hourly values of system energy- (KWh)

3. Daily and weekly system energy (KW).

As far as randomly variable components are concerned the factors that give

rise to these, summarized as the meteorological conditions, cause large variation

in this aggregated load. These are typically weather related components like

temperature, cloud cover, and humidity etc & special effects like holidays. The

details are discussed in chapter-6

Process data is used during development of system models and therefore

the resulting models' quality is largely dependent on data integrity. For hourly

54

forecasts, there should be 3 to 5 years of hourly historical data for each electric

system or buss to be forecasted. In the suggested model, current, historical, and

forecasted time series data of above two components has to include.

5.2 Data collection & analysis of data.

5.2.1: Data collection for base & randomly variable components

Obtaining Weather Forecasts is an essential input to the load forecasting

process in the pre-dispatch and short-term periods. For the purposes of load

forecasting each zonal Supervisory Control and Data Acquisition (SCADA offices

will use the half hourly base weather fore cast by Department Of Weather

Forecast. Also past data on holidays & the effect on the base components is

available with the SCADA zonal offices.

Base components data is collected through the Supervisory Control and

Data Acquisition (SCADA) system, which provides analog measurements and

status information on a real-time basis to support operational needs. The base

component values are scanned for every 12. All the information is used for short

term demand forecasting.

5.2.2 ANALYSIS OF DATA

Following procedures are used for analysis of data

5.2.2.1:Calculation of Short Term 10% and 90% POE scaling factors

Scaling factors applied to short-term load forecasts will be determined on the basis

of statistical analysis of historical load forecasting performance. The standard

55

deviation of the forecast error is calculated and the scaling factors are determined

by applying following equation

Scaling Factor = 1 ± (1.28 * se / Average Load) Where:

se is the standard deviation of the forecast error for the 50% POE forecast.

re corrected prior to establishing the relationships.

Assume the same relationship for each week within a month. Use these

relationships to scale the 50% and 10% among the days of the week. Refer

appendix A for details.

5.2.2.2 Treatment of Load Variations Due to Special Events In the Year

Special periods such as Diwali holidays etc. have significant effects on the zonal

load. For example significantly different daily peak load patterns are experienced if

Diwali day falls on a weekday compared to a weekend. To factor this phenomenon

into the medium term forecasts, historical data (temperature correction may be

necessary) will be searched to find load patterns (relationship of the peak loads)

within the week for example if Diwali falls on a Sunday compared to if Diwali falls

on a weekday. Store these relationships in a suitable form (normalized indexes)

and use them when the daily peak loads are derived using weekly peaks.

56

CHAPTER-6

SHORT TERM DEMAND

FORECASTING MODEL

57

6.1 Introduction

At any one time, to maintain a particular voltage throughout electricity grid

the amount of electricity drawn from the grid and the amount generated should

balance. This figure is usually called the electricity demand and, in the absence of

blackouts or demand shedding, is equivalent to the demand for electricity. It has

long been known that electricity demand is highly predictable as a result of its very

strong daily, weekly, and yearly periodic behavior. For example, figure 6.1 below

plots the half-hourly electricity demand for the B.S.E.S for four weeks in 2003.

Apart from the strong periodicity, one thing that can be noticed is that the “demand

profiles” also vary across season. Most short-term forecasters address this

problem by using only a two or three week moving window of data to estimate

their model, over which the periodic profile is effectively static.

Short-term Load forecasts are usually required on a half-hourly basis for

various time horizons. The load for a given area consists of various deterministic

components reflecting industrial, commercial and residential energy requirements

58

and a stochastic component. This latter weather-related load is governed both by

current conditions (e.g. temperature). The relationships with meteorological

variables can be quite non-linear (e.g. with thresholds) and may produce

unexpected changes and threshold responses. Short-term loads are affected by

human activities, recent weather conditions and the intervention of utility network

operators seeking to optimize distribution for a variety of customers.

Anthropogenic activities have strong diurnal, weekly and annual characteristics

(i.e. strong statistical seasonality) that produce most of the load variability. A

smaller component (variously estimated at 5-20%) in load demand is related to the

meteorological conditions, with maximum demand in B.S.E.S now having shifted

to extreme temperatures during summer daytime (high air-conditioning loads),

rather than cool winter mornings and evenings (when heating demands are the

greatest). Public holidays (national, state and local) can produce demand profiles

much closer to those experienced on weekends. Interruptible loads (e.g. off-peak

hot water heating) have been the main demand-side management influences in

past load histories.

As Electricity demand is influenced not only by temperature but also wind

speed, humidity, precipitation, evaporation, and cloud cover. These influence air-

conditioning, space heating, refrigeration and water pumping loads, which add to

the peak and 24- hour demand. The peak loading is particularly important, since

on occasions of extreme temperatures this is likely to stress electricity systems in

meeting demand. The impact on other uses of electricity can be significant: water-

pumping requirements will increase where the climate change becomes warmer

but not wetter, due to higher water demand from residential, commercial and

municipal sectors. Refrigeration requirements would increase and water-heating

59

requirements would decrease, although the direct effects are likely to be

significantly less than the effects on space conditioning. Refrigeration and water

heating equipment is often located in conditioned spaces and thus is not affected

by outdoor temperature changes. So while developing a model or procedures it is

imperative to take theses parameters in-to consideration. In this chapter of the

project the pre-dispatch & short-term procedures are represented & medium term

forecasting model is proposed.

6.1 The half-hourly electricity demand

60

6.2 Forecasting for B.S.E.S distribution area

Any practical forecasting scheme should make full use of the strongly

periodic and/or predictable components of the load before attempting to

investigate the more stochastic elements. A comprehensive review of the literature

and available commercial schemes reveals the dominant methodologies as load

profile analysis, pattern recognition, artificial neural networks (ANN) and a variety

of time series approaches. ANN methods have recently had considerable success

but are often criticized for their lack of transparency and shift of emphasis to the

training of the network. The demand forecasting procedure utilized in this project

uses a computer software for hourly load short ...medium & has a built in best

sides of more formal approaches like ARIMA, regression analysis and load type

61

libraries while at the same time avoiding their weak sides. It can be used

interactively or automated.

6.2.1OVERVIEW OF THE FORECASTING METHOD

As the zonal load level at a given time is influenced by a number of major

deterministic & random or stochastic factors. Log transformation is considered

essential due to heteroscedasticity and the following may summarize other major

effects: -

1. Trend effects - long term changes due mainly to economic growth,

domestic population growth and commercial and industrial growth patterns.

2. Cyclic time effects - seasonal effects such as yearly weather seasonality,

weekly anthropogenic patterns and diurnal cycles.

3. Special effects - public holidays, strikes and power downs.

4. Weather effects - various meteorological factors including temperature and

humidity.

5. Random effects.

These five effects may be separated into two categories, deterministic and

stochastic. Effects (2) and (3) are classed as deterministic and (1), (4) and (5)

classed as stochastic. For short term forecasting we will take into consideration

the weather & special day factor into the model considering the high correlation of

base components with these factors. For effect of special events that is public

holidays, strikes, power downs, different profiles were obtained for public holidays,

with two types being distinguished - the public holidays around the Diwali and New

Year time and other public holidays. Strikes and power downs should be included

as much as the available data will allow. Fig 6.2 gives the scheme for short-term

62

forecast & 6.3 gives the process of short-term load forecasting factors. It is

essential that deterministic cycles and any special effects be removed as

completely as possible so that the remaining information in the data can be

modeled as stochastic variation, a medium term demand-forecasting model is

proposed which will filter these effects.

63

Figure 6.2: Short term Forecasting Process

64

Fig: 6.3 Short-term forecasting process algorithm.

65

6.3 Detached Regression Analysis

Short-term load forecasting provides load data for each hour and covers a

period of one week. The load data are typically: 

Hourly or half-hourly peak load (KW)

Hourly or half-hourly values of system energy- (KWh).

Daily and weekly system energy (KW).

The forecasts of the hourly load which may correspond to a system load, an

area load or a busload; are of direct use for the short-term operation of the

system. The data parameters transmitted by the measuring equipments installed

at the Pre-dispatch and short-term load forecasts will be produced on the basis of

calendar days from the half-hour ending 0030hrs to half-hour ending 2400hrs.

Accordingly, eight calendar days are required to be forecast to fully cover the

seven trading days of the short-term period. The Medium term load forecast will

begin from the eighth day ahead.

As discussed previously, the Bureau of Meteorology weather forecast

includes the maximum and minimum temperature and expected conditions for up

to seven days ahead. Each zonal load forecast is reviewed on the basis of the

most recent weather forecast and the current zonal load. A detached regression

analysis is used for that.

The forecasting method used is based on the usage of comparison days

from history. For each day to forecast a dynamic set of comparison day selection

66

criteria is created. The criteria take into consideration day-of-week, day type

including special holidays, week and weather definable parameters. Comparison

days are sought in an iterative fashion with widening search criteria until one or

more days are found. Outside temperature is a strong explaining factor.

Forecasting allows for the inclusion of one user chosen explaining factor, which

influence is automatically determined through a detached regression analysis

process. The explaining factor must be an hour time series like target. It is unto

the user to decide if the data input will be pure measurements, or for example, a

12 hours moving average. Regression analysis is done on history data and with

only one explaining factor. The following detached analysis module performs

linear regression analyses by using the least squares method to fit a line through a

set of observations As the best regression analysis results are achieved when only

those factors involved in the load variation are left variable & all other factors are

held constant, the day, week and season variations are eliminated by analyzing

each hour, day type and month separately.

Following method of detach regression is used.

67

.

68

Regression correction is thus possible only when we have the fore cast for

the explanting variable. The program gives an option to manually write the

forecasted values, in case they are not found in the database.

Figure 6.4 shows the comparison days selection method

69

Figure 6.4 The comparison days selection method

70

6.4 Forecast level Correction

Comparison days can be, and are, weeks old. But the level can change

fast, especially in spring and autumn. Level correction is based on scaling the day

energies of the comparison days to the day energy of the most recent measured

suitable day. The level correction is calculated for each comparison day

separately, taking into account that the nearest measured day can be of a different

day of week type.

71

6.5 Pre-dispatch forecasting methodology

The equivalent day load forecasting methodology aims to produce a load

forecast for a single day, which is based on historical data with manual adjustment

to incorporate other known factors. This method has been applied to the pre-

dispatch and short-term forecasts. The technique initially relies upon the selection

of a similar type of day to that which is being forecast. Once this has been

achieved the forecast values are modified to reflect known current conditions.

The resulting load forecast is regarded as being based on a combination of

historic data, weather forecasts and expected system conditions. The

methodology is a two-step process as follows:

1. Search historic load records for an equivalent day using a set of criteria

based on the forecast conditions. These criteria will include one or more of

the following.

Date specification – month/s, year/s and day type.

Maximum and minimum temperature.

Expected daily energy.

Expected peak load.

2. Manually adjust the equivalent day profile to reflect the expected conditions.

This should cater for any differences between the historical conditions and

the expected conditions of the forecast day.

The process is represented by the flowchart given in figure 6.5.

72

Figure 6.5: Pre-dispatch forecasting methodology

73

6.6 Building a Medium term forecasting model

As discussed earlier the zonal load level at a given time is influenced by a

number of major factors. Log transformation is considered essential due to

heteroscedasticity and other major effects, which are summarized, as

deterministic and stochastic. By systematically decomposing the data into different

components, each modeled in a particular way, a model procedure can be

obtained that supplies valuable forecasts as well as providing insights into the

underlying determinants of the load. There are three deterministic cycles that are

obvious in the load data. These are the daily cycle of hourly or half hourly values,

the weekly anthropogenic cyclic effect and the yearly seasonality. A smoothed

average yearly cycle can be removed from the data to form a de-seasonalized

data set. The weekly anthropogenic patterns and the diurnal cycles can then be

calculated and removed. Both of these profiles can be produced for each month of

the year, as the variation from month to month is sufficiently significant.

Another important source of variation is the effect of special events that is

public holidays, strikes, power downs and the switching of special hot water

customers from one time to another. Different profiles were obtained for public

holidays, with two types being distinguished - the public holidays around the Diwali

and New Year time and other public holidays. Strikes and power downs should be

included as much as the available data will allow. It is essential that these profiles

(deterministic cycles) and any special effects be removed as completely as

74

possible so that the remaining information in the data can be modeled as

stochastic variation.

After the removal of the profiles the next characteristic to be modeled is that

of the long-term growth, or the trend component of the data. The long-term

changes are due mainly to economic growth, domestic population growth and

commercial and industrial growth patterns. If this information is not readily

available, and if the area to be forecast is too coarse for detailed area growth

information to be useful, then the trend component can only be estimated from

within the series itself. The proposed model is as shown.

75

Figure 6.6 Medium term forecasting model process

76

6.7 Conclusion

77

The model worked satisfactory in terms of its adaptability for the changes in

the load behavior. The model is intended for use in many different cases. Updating

the forecast with new available data was possible with the model. The

acceptability of model depends upon the reliability, & even under in the

exceptional circumstances, it must not give rise to unreasonable forecasts. For the

proposed method treating the special holiday periods of the year, substantial

amount of historical data will be required; otherwise the special holiday periods

may have to be hand-dressed

Most important was that it could be extended for the medium term

forecasting. Once the medium term forecasting model is developed, the model

needs to be tested for all zones and fine-tuned.

78

CHAPTER- 7

DISCUSSION OF RESULTS

79

7.1: Introduction

The results show that forecasts for the first day are not always better than

for later days. Because of the forecast approach of using comparison day,

forecasting exactly at midnight so-to-speak is the only time where the short-term

forecast error correction often misbehaves. Especially when the first hour is of a

different day type as the last measurement, as in both April and July, the short-

term error corrections are a hazard. January 26th is a Saturday. Regression

coefficients for Saturdays in January were automatically dropped, because they

didn’t meet the preset goodness criteria, a consequence of too similar

temperatures for previous Saturdays in January. Thus temperature corrections are

left out, which can be noticed in the higher MAPE for days 4&5. The results show

high co-relation between the temperature & demand. The results are discussed in

detail in section 7.2

80

7.2: Some test results

Figure 7.1 gives the graphical representation of the result.

81

Figure 7.1 Test results

82

Figure 7.1continued

83

The short term forecasting method produced reliable forecasts of electricity

demand. To illustrate the sensitivity of demand to temperature changes, the

demand model in detached regression analysis was driven by uniformly raised

temperatures in steps of 1°C up to 4°C. Figure 7.1 show the effect. It can be seen

that there are significant shifts in demand over the day. The 4°C rise increases

demand by 140 MW – equivalent to 2.6%/°C at system peak demand. With this

simple linear model, each hourly demand is raised by an equal MW amount and

clearly does not indicate the relative sensitivity of different periods of the day (e.g.

air conditioning use at night versus early afternoon). A more sophisticated

approach will be necessary to fully identify the changes in daily load profiles that

will result from rising temperatures. Furthermore, it is anticipated that improved

demand models will be gained using additional weather variables that are known

to influence demand as well as increased use of prior demand and temperature

measures. Once the variables determining the short-term demand forecasting

were determined the expected benefit was in term of determining the process flow

for medium term forecasting model. One future refinement to the medium term

model could be a more precise modeling of the temperature effect. In the short-

term model, the hourly temperature values could not be properly utilized, so the

best results were obtained when using only daily average temperatures. The

reason for this seems to be that the changes in the temperature are slow, and the

succeeding measurements are correlated even with quite long delays. Another

possible approach would be to try to separate the temperature effect, and forecast

the temperature-normalized load. The temperature dependent part would then be

added to this in order to get the final load forecast.

84

Figure7.2: Demand With uniform temperature rise

85

7.3 Precautions

In B.S.E.S the long-term demand forecasting is done by using trending

method, which uses the values of units sold for forecasting. The sort-term &

proposed long term forecasting models will need two-dimensional approach to

tackle the problem. One alternative for existing system would be to rebuild the

current demand forecasting system. This would require two major efforts. First, the

zonal region would need to collect new data on current uses of electricity and end-

use level. Second, new forecasting models, which can integrate the existing

trending model, would have to be developed. Clearly, this approach would result in

forecasts that provide all of the details and information required by the proposed

system. The costs would be very high. Collection of the underlying data on energy

use in the zonal would be an expensive affaire. The cost of collecting such data

would probably be over 0.25 cr. Analysis of the data and incorporating into a set of

demand forecasting models would be comparable in cost to the data collection.

Building an integrated set of zonal l demand forecasting models, electricity price

model and a load shape forecasting system with an efficient demand forecasting

system would another 0.25 cr. These are very rough cost estimates, but a total of

over 0.50 cr, may be in the ballpark.

86

7.4 Conclusion

The result has clearly shown high correlation between temperature & load.

Further the short term forecasting method produced reliable forecasts of electricity

demand. The MAPE was in the acceptable limits & the pilot short-term model can

be further developed for the proposed medium term model. The proposed model

uses regression method & further extension of this model for long-term forecasting

will be a costly affair as the current long-term model is based on trending method.

87

CHAPTER-8

FUTURE SCOPE

88

8.1 SUMMARY

The motivation for accurate forecasts lies in the nature of electricity as a

commodity and trading article; electricity cannot be stored, which means that for

an electric utility, the estimate of the future demand is necessary in managing the

production and purchasing in an economically reasonable way. In India, the

electricity markets have recently opened, which is increasing the competition in

the field. It gives scope for the utility firm to concentrate on short term forecasting.

This work concentrates on short-term forecasting, where the prediction time varies

between a few hours and about one week. The short-term model uses the

detached regression analysis. The variables used for this model are basic

electricity variables (KWH & KW) & random components e.g. temperature &

holydays. The model is reliable in terms handling the extreme values of load &

suffices the following objectives of building the model.

The model is automatic and able to adapt quickly to changes in the load

behavior.

The model is intended for use in many different cases.

Updating the forecast with new available data is possible.

The hours closest to the forecasting time is forecasted as accurately as

possible.

The model is reliable. Even exceptional circumstances it does not give rise

to unreasonable forecasts.

Difficult weather conditions typical in supply area of B.S.E.S, especially

large variation of outdoors temperature, are taken care of.

89

The model is easily attachable to an energy management system.

8.2 Future Scope

One future refinement to the model could be a more precise modeling of

the temperature effect. The hourly temperature values could not be properly

utilized so the best results were obtained when using only daily average

temperatures. The reason for this seems to be that the changes in the

temperature are slow, and the succeeding measurements are correlated even with

quite long delays. It is interesting to see the effect of humidity factor on load,

especially in summer & in month of October.

One concept to consider for a future study, as far the medium term model is

concern is the impact age has upon residential demand. An interesting idea would

be to study the impact school age children have upon summer peak demand.

Children are home from school all day during the summer. This means that

electrical appliances such as televisions, computers, and radios are on all day,

and rather than turning the air conditioning down when someone is not home, as

may people do, the air conditioning would have to remain on at a high-level to

remain comfortable for the children. Therefore, with children are home during the

summer for the hottest portion of the day the residential summer peak may be

affected.

A future study could analyze the impact of the migration from the urban to

rural areas that is taking place in Mumbai. Therefore, studying the impact of the

migration from rural to urban may help forecasters determine what can be done to

meet the changing needs of West Virginia’s consumers.

90

List of References

Ackerman, Gary. (1985). Short-Term Load Prediction for Electric-Utility Control of

Generating Units.

In D.W. Bunn and E.D. Farmer (Eds.), Comparative Models for Electrical Load

Forecasting. (pp. 33-42). New York: John Wiley & Sons Ltd..

Al-Alawi, Saleh M & Syed M. Islam. (1996). Principles of Electricity Demand

Forecasting: Part 1 Methodologies. Power Engineering Journal,10 (3), 139-143.

Al-Garni, Ahmed Z., Syed M. Zubair & Javeed S. Nizami. (1994). A Regression

Model for Electric-Energy-Consumption Forecasting in Eastern Saudi Arabia.

Energy, 19 (10), 1043-1049.

Anderson, David R., Dennis J. Sweeney, & Thomas A. Williams. (1978).

Essentials of Management Science: Applications to Decision Making. St. Paul,

Minn: West Publishing Co. Bartels , Robert & Denzil G. Fiebig. (1996).

Metering and Modeling Residential End-Use Electricity Load Curves. Journal of

Forecasting, 15, 415-426. Baum, Vladimir. (1993). Glowing in the Gloom.

Petroleum Economist, 60, 6-9. Bureau of Business and Economic Research.

Consumer Price Index-All Urban Consumers, West Virginia, (1980-1995),

wvbeis.be.wvu.edu/public/cat/cpi80p.txt.

Burns, Thomas G. (1984). The Art and Science of Energy Forecasting. Journal of

Petroleum Technology, 36, 1437-1442.

Charles River Associates. (1986). A Guide to Electricity Forecasting

methodology. Washington D.C.: Edison Electric Institute.

Cowing, Thomas G. & Daniel L. McFadden. (1984). Microeconomic Modeling and

Policy Analysis: Studies in Residential Energy Demand. Orlando, Fla.: Academic

Press, Inc., Harcourt Brace, Jovanovich Publishers.

Crow, Robert T., Michael Robinson, and Raymond L. Squitieri. (1981). Forecasting

Electricity Sales and Loads: A Handbook for Small Utilities.

Donnelly, William A. (1987). The Econometrics of Energy Demand: A Survey of

Applications. New York: Praeger.

Granger, C.W.J. & Paul Newbold. (1986). Forecasting Economic Time Series

Academic Press, Harcourt Brace Jonavich Publishers.

Griffiths, William E., R. Carter Hill, & George G. Judge. (1993). Learning and

Practicing Econometrics. New York: John Wiley & Sons, Inc. 100

91

Gupta, P.C. (1985). Adaptive Short-term Forecasting of Hourly Loads Using

Weather Information.

In D.W. Bunn and E.D. Farmer, Comparative Models for Electrical Load

Forecasting. (pp. 43-56) New York: John Wiley & Sons Ltd.

Halvorsen, Robert. (1970). Demand for Electric Energy in the United States.

Southern Economic Journal, 42, 610-625.

Halvorsen, Robert. (1978). Econometric Models of U.S. Energy Demand.

Lexington, Mass.: Lexington Books, D.C. Heath and Company.

Harvey, Andrew C. (1983). Time Series Models (2 nd ed.). Cambridge, Mass.: The

MIT Press.

Houthakker, Hendrikk S. (1980). Residential Electricity Revisited.

92

APPENDIX: A

Description of Forecast at 10%, 50%, 90% POE Components

Some statistical terms are commonly used with regard to NEMMCO load forecasting.

These terms are:

50% POE: 50% probability of exceedance. This is where 50% of the actual

load values are expected to be above the forecast and 50% of the actual load

values are expected to be below the forecast. This the most likely load as

predicted by the person producing the forecast. In the Short Term period the

50% probability of exceedance forecast is based on historical demands and

the weather forecast.

10% POE: 10% probability of exceedance. This is where 10% of the actual

load values are expected to be above the forecast and 90% of the actual load

values are expected to be below the forecast. This is the load that against

which reserve criteria will be assessed.

90% POE: 90% probability of exceedance. This is where 90% of the actual

load values are expected to be above the forecast and 10% of the actual load

values are expected to be below the forecast. This is the load used in

assessing minimum load conditions.

The use of these three components to describe a forecast results in a most likely

forecast load value, the 50% POE, and a forecast band, from 90% POE up to 10%

POE, where most of the actual load values are expected to fall. This can also be

described in terms of a mean forecast load value and an associated 80% confidence

interval. The load forecast for all regions will be based on a 50% POE profile. The

10% and 90% POE forecasts will be calculated from the 50% POE profile. Various

scaling factors will be implemented to take variations into consideration to improve the

forecast reliability.

Appendix 1.1 - Calculation of Short Term 10% and 90% POE scaling factors

Scaling factors applied to short-term load forecasts will be determined on the basis of

statistical analysis of historical load forecasting performance. The standard deviation

of the forecast error is calculated and the scaling factors are determined by applying

Equation 1.0.

93

Scaling Factor = 1 ± (1.28 * se / Average Load) (1.0)

se is the standard deviation of the forecast error for the 50% POE forecast.

Approximately 1.28 standard deviations either side of the mean represents the 80%

confidence interval determined by the 10% and 90% POE forecast.

Short Term 10% and 90% POE scaling factors

94

ACKNOWLDGEMENT

I take the opportunity to express my gratitude to Mr. J.D.Nerkar for his valuable

guidance in the project work. I also thank my team member how were part of the

project & help me in giving insight of the project.

95