JIMMA UNIVERSITY

Department of Electrical And Computer

Engineering

Power System Planning and Management:

ECEG-6306

By: Dr. Milkias B. (Ph.D.)

Asssistant Professor, Adama

Science And Technology University
Milkias B. 8/11/2020 1
POWER SYSTEM
PLANNING
AND
MANAGEMENT-
ECEG-6306
Milkias B. 8/11/2020 2
 OUTLINE
▪ 1. FUNDAMENTALS
 Introduction
 Planning & economics
▪ 2. Load data and Forecasting
▪ 3. Power quality and Reliability analysis
▪ 4. DESIGN PRINCIPLES & PRACTICE OF SUBSTATION
 Substation sizing and spacing
 Substation location
 Effect of changing load density
 Effect of changing primary voltage
 Cost interaction of substation size and spacing vs primary voltage and
load density
▪ 5. PROJECT MANAGEMENT
 Project evaluation
 Financing
 Project phases
 Key points
▪ 6. UNIT COMMITMENT
 Unit commitment
 Spinning reserve
 Priority list method/Merit order scheduling
Milkias B. 8/11/2020 3
 2.
LOAD DATA AND
FORECASTING

Milkias B. 8/11/2020 4
Lecture Outline:
▪ Classification and Characteristics of Loads

▪ Load Curves and Load duration curves

▪ Forecast Methodologies

▪ Demand forecasting

▪ Peak Demand Capacity Forecasting

Milkias B. 8/11/2020 5
 2.1 LOAD DATA AND FORECASTING
A/ Load Data:
▪ The load data is needed for defining the requirements of the network’s
transmission capacity, approximating the transmission losses or estimating the
existing network’s capability to transfer increasing loads.
▪ The planning of new generation capacity or energy purchase requires
knowledge of customers’ load variation.
▪ The manner in which load curve data are collected, recorded, analyzed, and
represented can produce a dramatic effect on what the resulting load curves
look like and the perceived values of peak load and coincidence.

▪ Sampling rate and sampling method both have a big impact on what load curve
data look like, how accurate they are, and how appropriate they are for various
planning purposes.

Milkias B. 8/11/2020 6
▪ The load data may be formulated in several ways according to the
requirements of applications. The most important specifications for load
data are
 System location: customer site, low voltage network, transformer etc

 Customer class: industry, service, residential, electric heating, etc

 Time: time of year, day of week, time of day

 Time resolution of the load recording: 5min, 15min, 30 min etc

▪ Load behaviour at the distribution level is dominated by individual

appliance characteristics and coincidence - the fact that all customers do not
demand their peak use of electricity at precisely the same time.

Milkias B. 8/11/2020 7
▪ Suppose one were to consider one hundred homes, for example one hundred
houses served by the same segment of a distribution feeder. Every one of
these homes is full of equipment that individually cycles on and off
appliances like - water heaters, air conditioners, heaters, and refrigerators
that all are controlled by thermostats.
▪ Thus, one particular household might peak at 22 kVA between 7:38 AM and 7:41 AM, while
another peaks at 21 kVA between 7:53 AM and 8:06 AM, while another peaks at 23 kVA
between 9:54 AM and 10:02 AM. These individual peaks are not additive because they
occur at different times. The individual peaks, all quite short, do not all occur
simultaneously.

Milkias B. 8/11/2020 8
▪ Peak load per customer drops as more customers are added to a group. Each
household has a brief, but very high, peak load - up to 22 kW in this example for a
southern utility with heavy air-conditioning loads.

▪ This tendency of observed peak load per customer to drop as the size of the customer
group being observed increases is termed coincidence and is measured by the
coincidence factor, the fraction of its individual peak that each customer contributes
to the group's peak.

Milkias B. 8/11/2020 9
▪ The diversity factor, D, measures how much higher the customer's

Individual peak is than its contribution to group peak. D = 1/C

▪ As customers are combined in groups, as when the planner analyzes load for
groups served by equipment such as service transformers, laterals, and
feeders, the erratic load curves add together with the sharp peaks
intermingling and forming a smoother curve.

Q.3 Determine the peak load of transformer supplying to 30 households having

the peak demand of 15kw. C(15) = 0.48.

Milkias B. 8/11/2020 10
 …
Load refers to customer demand for electricity. This section
categorizes the different load types and estimates the future load that a
power system will supply sufficient and economic energy to the
customers’ electricity demand. Power system planning starts with a
forecast of anticipated future load requirements. Estimates of both
demand capacity and energy requirements are crucial to effective
system planning.
Why Load Forecasting is important in Power System?
• There is a time lag between awareness of a future need and serving that
need.
• In order to serve the need, adequate generation, transmission and
distribution system have to be put in place. But delivery of requested
power and energy always requires some time for the construction of the
generation and transmission facilities which takes a definite lead time.
• Thus, the time lag is the main reason for forecasting and planning.

Milkias B. 8/11/2020 11
…

Milkias B. 8/11/2020 12
 Characteristics of Loads
• In most part of the world, the residential loads have the most seasonal
fluctuations. And they are responsible for the seasonal variations of the
system peak. This is mostly due to the weather sensitive devices used
for space heaters and air conditioners.
• In the case of Ethiopia, the weather does not significantly vary
throughout the year. Therefore demand seasonality is minimal. Here
the residential loads are characterized by lighting and since the
system peak is during the evening, their contribution to the system
peak is significant.

Milkias B. 8/11/2020 13
 …
• Again in some parts of the world, commercial and public services loads
are characterized by seasonal fluctuations just like that of the domestic
category.
• In the case of Ethiopia there is no significant demand seasonality.
• The Addis Ababa case is very much different. It is probable that the
composition of the load is also very different as the Addis Ababa
commercial and public services load includes many large institutions
and schools working mainly between 8.00 am and 6.00 pm.
• the Addis Ababa suburbs would largely consist of small shops that stay
open throughout the evening as well as hotels, bars, cafes and
restaurants with a strong evening trade.

Milkias B. 8/11/2020 14
 …
• Industrial loads are considered base load that contain little weather
dependent variations. The same is true for the Ethiopian case. The
industrial load peak is during the day. Due to shift work most of the load
continues till midnight.

Milkias B. 8/11/2020 15
 2.3 Load Curves and Load duration curves

Load curve is a graphical representation between load in kW (or MW) in proper

time sequence and time in hours. It shows the variation of load on the power
station.
▪ The highest point on the load curve represents the maximum
demand (peak demand) on the station on that day.
▪ The area under the load curve gives the total energy supplied on
that day.
▪ if this area is divided with the number of hours gives the average
load (power) on the station.
• Load curves can be daily, weekly, monthly and yearly depending on the
period of observation. For daily load curve, the period of time is taken as 24
hours and for annual load curve the time is considered as 8760 hours.

Milkias B. 8/11/2020 16
 Load Factor
Load factor of a system is the ratio of the average load over a given period of
time to the maximum demand (peak load) occurring in that period.
Load Factor=Average Load/Peak Load
❖ It would be ideal to have a flat load curve. In practice, load curved are far
from flat. For a flat load curve, the load factor will be higher.
❖ Higher load factor means more uniform load pattern with less variations in
load. This is desirable from the point view of maximum utilization of
associated equipment which is selected on the basis of maximum demand.
❖ Ethiopia’s power system has a load factor of around 0.57 (in 2006).
The industrialized countries have a load factor higher than this.

❖ This means, the power system operates at an average 57% of the

peak system load in the year 2006.

Milkias B. 8/11/2020 17
System load duration curve for the year 2006 of the Ethiopian power system

Milkias B. 8/11/2020 18
 …
100
95
90 Peak Load
85 Plants
80
75
Percent of Peak Load

70
65 Intermediate Load
60 Plants
55
50
45
40
35
30
Baseload
25 Plants
20
15 Capacity (MW) *
10
5
0
0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 4,500 5,000 5,500 6,000 6,500 7,000 7,500 8,000 8,500
Cumulative Hours

Milkias B. 8/11/2020 19
example
 HW
A system with a single power plant has the following daily load variation.

Time in 6-8 8-11 11-16 16-19 19-22 22-23 23-00 00-6

Hour

Load in 30 40 50 35 70 65 55 20
MW

Question:
▪ Draw the daily load curve
▪ Determine the maximum demand
▪ Determine the average load
▪ Determine the load factor
▪ Draw the load duration curve

Milkias B. 8/11/2020 22
example
 Load
Forecasting

Milkias B. 8/11/2020 24
 Objectives of Load Forecasting

▪ It determines the future trend of electricity consumption of each

demand sectors starting from the base year for the whole projection
period.
▪ It determines the total annual consumption as well as its discissions in
to local and interconnection power pool loads.
▪ It depends mainly on the historical and current data of the base year,
national GDP growth, population growth, customers’ income growth,
government plan on rural electrification, government policies and view
on climate issues etc.
▪ It includes the historical data analysis and modelling annual forecast
of energy demand and peak capacity demand of a power system
network within the projection period.
Milkias B. 8/11/2020 25
 B/ Load Forecasting and Utility Planning:

 Load forecasts are the starting point for the entire chain of power delivery.

 Planners require a load forecast before they can set an optimal dispatch. The
load forecast must be of sufficient detail to provide necessary information to
the dispatch model.

 Usually all the needed load data is not available directly and the load values
must be estimated and forecasted using other available information.

 The load modelling and forecasting is based on knowledge of several factors

influencing the customer’s load. The most important factors are:
 Population: overall residential demand and also commercial and industrial activity
increase with increasing population.

 GDP: commercial and industrial load is increased or decreased by strong or adverse

economic growth. In addition, the wealth of households is also highly correlated with
demand.
Milkias B. 8/11/2020 26
  Temporal behaviour patters: electric consumption varies over time cycles in relation to
variation in human activities. Like vacation, school cycles, weekday and weekend,
holidays, and also early morning hours, work hours, evening hours and sleeping hours.

 Weather parameters: it follows seasonal and daily patterns, which coupled with
temporal behaviour impact on residential, commercial and industrial load. This
includes temperature, humidity, sunshine and cloudiness, etc

▪ A load shape is the power requirement as a function of time over a given time
interval. There are daily, weekly, and seasonal load patterns.

▪ Load predictions are made over different time horizons. Based on this, future
load demand can be estimated in either of the following conditions:

Milkias B. 8/11/2020 27
▪ Short-term forecasts of 1 to 2 years are mainly of value in deciding operating
procedures and preparing budget estimation.
▪ Short term forecast indicates the sales and purchase of the power.
▪ Short term forecast indicates the development of distribution networks.
▪ Short term forcast is required for scheduling a next-day dispatch. i.e, it
includes forecasting day ahead electricity demand, available supply and
ancillary services and setting day ahead dispatch schedule.
▪ Once a dispatch schedule has been set, planners forward it to day-ahead
schedulers. There are generation schedulers, inter-utility schedulers, fuel
schedulers, and transmission schedulers.

▪ The generation scheduler has the responsibility of informing each generation

facility of its next day’s operating schedule.

▪ Inter-utility contract schedulers inform their utility counterparty of required

next-day power exchanges.
Milkias B. 8/11/2020 28
▪ Fuel schedulers inform their suppliers of next-day fuel deliveries.

▪ Transmission schedulers schedule power flow exchanges across

transmission lines of neighbouring control areas.

▪ This technique does not depend on population and GDP; however, it is

sensitive to temporal behaviour and weather.

▪ For day-ahead scheduling hourly forecasts are necessary, although some

planners use sub-hourly forecasts.

▪ An hourly forecast is established as an energy requirement in MWH or a

power requirement in MW.

Milkias B. 8/11/2020 29
 Mid-term Load Forecasting and Utility Planning
▪ Medium-term forecasts time period varies from 5 to 6 years of planning and
size of the power station.
▪ Medium term forecast indicates the transmission and distribution losses.
▪ Medium term forecast indicates the sales and purchase of the energy.
▪ Medium term forecast indicates the Energy conservation.

• As with short-term planning, load is the starting point of mid term planning
because plans are set to supply load.
• Mid-term plans must be set to meet many possible outcomes for the load, so
planners must develop more than a single load forecast depending weather
condition, population growth, etc.

Milkias B. 8/11/2020 30
▪ It also optimizes dispatch costs over a time horizon ranging between 1 month
and 3 years.

▪ Mid-term planning address issues related to maintenance scheduling for both

generation and transmission, use of energy-limited resources, and capacity
contracting with other utilities.

▪ One aspect of midterm utility planning is support of decision making for

capacity transaction.

▪ In short, midterm planning is the decision to include or exclude a

resource from a utility’s capacity base in accordance with maintenance
requirements, weather and environmental limitations, or the price and
availability of capacity with in a network of utilities.

Milkias B. 8/11/2020 31
(1) Long term forecast
▪ Long-term forecasts time period varies from 15 to 20 years of studying the
energy problems.
▪ It takes four to six years for the construction, installation and maintenance of
the equipment in power-stations.
▪ Long term forecast indicates the sales and purchase of the equipment.
▪ Long term forecast indicates the energy policies.

▪ Long-term forecasting addresses the annual changes in demand that reflect

changes in population and GDP. Although weather is a major driver for

short-term forecasting with an explicit role, it plays a background role in

long-term forecasting.

Milkias B. 8/11/2020 32
▪ Planners must have recommendations ready for project developers to
complete construction in time to meet customer load requirements.

▪ Project development encompasses the process of constructing a new

generating facility or transmission equipment from the conceptual phase
until the moment when the facility is operating.

▪ Aside from meeting the requirements of a baseline forecast, timelines are

also critical for contingency planning. Actual demands may vary from
baseline forecast, and planning must address this contingency.

Milkias B. 8/11/2020 33
 Forecasting Load and Energy Requirements

▪ The following methods are used for forecasts or estimates of future demand
of electrical energy
1. Load-survey methods
2. Methods of extrapolation
3. Mathematical methods
4. Energy consumptions.

▪ (1) Load-survey methods

▪ In this method, existing and future load requirements are forecast for the area
under consideration.
▪ The consumer groups are classified into residential consumers, domestic
consumers, commercial consumers and industrial consumers.
▪ This method consider the load forecast of next four to six years for the loads
increment in the various categories.
▪ This method is consider the characteristics of the load
▪ Load survey method distribution loss is 10 to 15%.
Milkias B. 8/11/2020 34
 (2) Methods of extrapolation
▪ This method involves the study of the data collected from various records
and is used in the comparison of trends of increase in demand and energy
consumption during past periods comparable with the periods of the
forecasts.
▪ Extrapolation methods rates are then used for the future estimates of the
forecasts.
▪ In the statistical method of prediction of future load and future possible
energy consumption, the curve of load (kW) against years and the curve of
energy consumed (kWh) against years should be plotted.
(3) Mathematical methods
▪ This method curves are plotted with the data of the energy consumption for
past years.
▪ This method uses the linear and exponential curve fitting techniques.
▪ This method is applicable where adequate statistical data over a
sufficient long period on the past consumption is available.

Milkias B. 8/11/2020 35
 2.4 Load Forecast Methods

• The accuracy of load forecasting depends not only on the load

forecasting techniques, but also on the accuracy of forecasted
weather scenarios.
• Various weather variables could be considered for load forecasting.
Temperature and humidity are the most commonly used load predictors.
Weather forecasting is an important topic while it is outside of the scope of
this course.

The key parameters during load forecast are population, gross domestic
product (GDP), temporal behaviour patterns, and various weather
parameters.

Milkias B. 8/11/2020 36
 …
Population Although energy consumption varies from household to household,
overall residential load is highly correlated with population. It stands to reason
that more households create more demand.
GDP The wealth of households is also highly correlated with demand. Wealthier
households have central air conditioning, larger living space, larger refrigerators,
and more appliances. They are less sensitive to utility bills and accordingly less
likely to economize their discretionary consumption, for example, turning up the
thermostat on central air conditioning. In economic upturns, residential
demand per household increases
Weather Parameters Weather has the largest impact on load. Weather follows
seasonal and daily patterns, which coupled with temporal behaviour impact
residential load. Below we indicate the more significant weather parameters
and discuss their impact on load.

Other like holidays, national days etc

Milkias B. 8/11/2020 37
 …
• In most cases the choice of a method will depend more on the background
and time available from the planning staff than on the technical merits of
the method.

• There are numerous techniques for modelling and forecasting electrical

energy and load requirements. Each analyst and organization is apt to
characterize its technique in a different way.

• The following scheme of classifying models emphasizes the different ways

in which analytical techniques treat customer choice and behaviour.
• Models are divided into: time series, econometric and end-use methods.

• The techniques may be based on extrapolation or on correlation or a

combination of both.

Milkias B. 8/11/2020 38
 A. Time series Regression (Extrapolation)

• Searching for systematic and recurrent relationships between

loads at various points in time is the essence of time series
modelling. Since time is the only explanatory variable, the
data collection requirements of this technique are the least of
any forecasting technique. However, not all time series
models are easy to implement.

• The predictive accuracy of time series analysis varies

tremendously from application to application.
• For short-term forecasts of relatively large and stable systems,
it can be as accurate as more elaborate models.
• It is subjected to extreme errors in the 10—20 year horizon
required for capacity expansion planning.
Milkias B. 8/11/2020 39
 …
• Extrapolation method makes the assumption that past rate of change in
electricity use or in electricity use per customer will continue into the
future.

• In mathematics, extrapolation is the process of constructing new data

points. It is similar to the process of interpolation, which constructs new
points between known points, but the results of extrapolations are often
less meaningful, and are subject to greater uncertainty.
• Extrapolation method makes the assumption that past rate of change in
electricity use or in electricity use per customer will continue into the future.

• Advantages of extrapolation method:

▪ It is the simplest type of forecast

▪ It requires only access to basic electricity sales and peak statistics
▪ It can be developed by the use of simple statistical models

Milkias B. 8/11/2020 40
 Disadvantages of Extrapolation Method
▪ They assume, generally that the future will be like the past, which often
turns out to be untrue.
▪ Changes in technstructural shift in the economy or in demography and
changes in regulations are just a few of the parameters that are extremely
difficult to capture with the extrapolation forecast.
▪ ology, Extrapolation method is primarily used for short term forecast (one to
five years), for which the assumption that the future will be like the past is
logical.
▪ The essential prerequisite for a time series forecasting technique is data for
the last 10 to 30 time periods.
▪ Three different forms of equations which uses the time series regression
analysis:
• Linear equation
• Demand=a*Year+b
• Parabolic equation
• Demand=a*year^2+b*Year+c
• Polynomial Equation
• Demand=aYears^3+bYears^2+cyears+d
Milkias B. 8/11/2020 41
B. Econometric Method (Correlation Techniques)

• This class of model, like the time series model, uses historical regularities to
predict the future but attempts to go beyond time series models in explaining
the causes of trends. Thus, Econometric Method depends on the relationships
between electricity use or peak demand and various economic demographic
variables.
• Econometric models postulate explicit causal relationships between the
dependent variable (either energy demand or loads) and other economic,
technological or demographic variables. Simple univariate models which use
the relationship between energy growth and GDP are in a sense causal since
they postulate, at least implicitly, that economic activity creates the need for
electricity.
• Other things being equal, econometric modelling would be preferred to time
series analysis.
• Even if both techniques could predict changes in demand with equal accuracy,
the econometric model would be more valuable since it might help in
understanding why changes in demand were occurring. Knowing causes can
help to plan to meet future needs.
Milkias B. 8/11/2020 42
 …
• In statistics, dependence refers to any statistical relationship between two
random variables or two sets of data.
• Correlation refers to any of a broad class of statistical relationships involving
dependence.

Milkias B. 8/11/2020 43
 …
Mostly used independent variables include:
•GDP, Population, Electrified Population, Price and GDP/Cap
The dependent variables
•Energy demand or Peak demand
Multiple Regression Equation
Y=f(x1, x2, x3, x4, x5)
=a+b*GDP+C*pop+d*electrified Population+e*GDP/Cap+f
Examples of economic and other variables often used to develop
econometric relationships are:
• Household income
• Electricity prices (by customer group)
• Labor productivity
• Commercial sector output (by sub sector)
• Price of other fuels
• Use of other fuels (gas,oil,coal,..etc)
• Employment (by sector and subsector)
• Industrial or agricultural productivity
Milkias B. 8/11/2020 44
 …
• Correlation techniques of forecasting relate system loads to
various demographic and economic factors. This approach is
advantageous in forcing the forecaster to understand clearly
the interrelationship between load growth patterns and other
measurable factors.
• The most obvious disadvantage, however, results from the
need to forecast demographic and economic factors, which
can be more difficult than forecasting system load.

Milkias B. 8/11/2020 45
 Econometric Regression Method
• In this method, both economic and demographic factors are considered to
be affecting the electrical energy consumption.
• Equations are developed for each sector to simulate the historical respective
energy consumption taking into consideration the proper economic and
demographic factors as independent variables.
• The factors considered can be GDP/sector, electricity price per/sector, and
population.
• The regression equation used for each consumption category is expected to
be according to the following equations:

E (i) = a * GDP(i) + b * pop(i) + c * price(i) + d

• When the mathematical equation is achieved, it is then used to predict the
future energy demand of each sector. Hence, the total annual energy
consumption prediction or forecast can be easily calculated by summing up
those of the different sectors. Finally the system generated energy and peak
load are derived by assuming the system losses and load factor.
Milkias B. 8/11/2020 46
 C. End use Approach

• End use forecast is different from time series regression and econometric
forecast in that they attempt to build up estimates of electricity needs starting
with analysis of what electricity will be used. This final class of model is more
diverse than the preceding types.

• In one form or another, end-use modelling is probably part of most utility

forecasting methods. Its distinguishing characteristic is a detailed description
of how energy is used. Such models usually begin by specifying reasonably
homogeneous uses for which energy is ultimately required, such as heating
water, cooling buildings and cooking food.
• The model then describes, via mathematical equations and accounting
identities, the types of energy-using equipment that businesses and
households have, and how much energy is used by each type of equipment to
satisfy the predetermined levels of end-use energy demanded.
• By summing up the units of equipment times the average energy used by each
class of equipment, total energy demand by fuel type is revealed.
Milkias B. 8/11/2020 47
 …
• The data requirements grow linearly and the computational difficulty grows
exponentially with the descriptive detail sought by the modeller.
• Optimization models not only require information on the conversion
efficiencies of energy-using technologies, they also require cost data.
• While the computational requirements of end-use accounting can be
satisfied with minimal computer time, optimization of realistic energy
models requires significantly more computing capabilities.
• End use approach used for long term demand forecast (20-30 years)
Advantages of End use Forecast:
▪ They can be quite detailed, providing more information for planners.
▪ They are not usually technically complex, requiring mostly simple arithmetic to
carry out
▪ Can provide integrated forecast of both energy and peak power demands
▪ The assumptions used in forecasting are usually relatively easy to follow, to check
and revise.
▪ End use forecast makes it more straight forward to estimate the impacts of energy
efficiency options and demand side management
▪ Useful for planning demand side management strategy (DSM, Motors, lambs)
Milkias B. 8/11/2020 48
 …
Disadvantages of End use Forecast:

▪ Hard to collect detailed data and time consuming to prepare.

▪ End use approach estimates total usage by forecasting small segments of
usage.
▪ End-use models are often weakest in predicting consumers' fuel-use
decisions. With the available data, they can easily describe where the
energy is being used and for what purposes but, without a theory to
explain choices, they are limited in their explanatory power to predict the
future. The ideal end-use model (rarely achieved) would, for example, not
only tell us the average watts of lighting energy in households, and how
this amount has changed over time, but also what caused households
and/or housing operators to make these changes.

Milkias B. 8/11/2020 49
 …
Examples of forecasting Methods
Simple Number 3,348

Growth Rate Growth(3%)

Growth(3%, 2010, 2%)

Interpolation: straight-line changes Interp(2000, 100, 2010, 120, 2020, 200)

between pairs of data years and values.

Step: discrete changes between pairs of Step(2000, 100, 2010, 120, 2020, 200)
data years and values.

Remainder: calculates remaining Remainder(100)

balance between parameter and values
of neighboring branches.

Milkias B. 8/11/2020 50
…

Milkias B. 8/11/2020 51
…

Milkias B. 8/11/2020 52
 2.5 Demand Capacity Forecasting

Demand capacity forecast may be designed to estimate the instantaneous

power (MW) that the system need at specific time with a given period or the
amount of electric energy (GWh) needed over the day week, month or year.

The accuracy of a forecast is crucial to any electric utility, since it dictates the
timing and characteristics of major system additions. A cost of having
insufficient electric generation capacity on hand to meet the customer’s needs
usually manifests itself quit visibly. Industrial facilities reduce production,
commercial establishments have interrupted hours of service, households
suffer inconvenience, goods may be damaged people may be injured.
Three characteristics of utilities set them apart from most industries and
heighten the negative consequences of in accurate expansion plans:
•They commit relatively large amount of capital for long periods,
•Comparatively long lead times are needed to add to their production capacity,
•They provide a critical input to the production process of many other
industries.
Milkias B. 8/11/2020 53
 …
Regression Analysis for Demand Forecast
Regression in general is a relationship between the variable that will be
forecasted (dependent) and another variable (independent). Simple regression
refers to a single independent variable. Multiple regressions refer to a multiple
independent variables.

Regression equations:
•1 Straight Line
•Y=a+bX
•2 Parabola
•Y=a+bx+cx^2
•3 Polynomial
•Y=a+bx+cx^2+dx^3
•4 Exponential
•Y=a+b^x

Milkias B. 8/11/2020 54
…

Milkias B. 8/11/2020 55
 …
Sectorial Modelling
The energy demand is usually forecasted using the above methods for each
category of demand (Domestic, commercial, industrial etc.). These have to be
aggregated in order to determine the total energy forecast.

Milkias B. 8/11/2020 56
 …
Example: Ethiopian Power System Demand Forecast
❖ Grid Electricity grows from 3,140 GWh in 2006 to just over 20,748 GWh in
2016,

Milkias B. 8/11/2020 57
 2.6 Peak Demand Forecasting

Peak demand forecast is important due to the seasonal variation of the capacity
demand by sectors. Thus a forecast that should analysis the peak capacity
demand of the customers is done usually by projection the peak demand of a
day, week, month or year using the different forecast methods until the
projection end year.
The maximum instantaneous load within a given utility service territory is
called its peak demand. In electric systems with predominantly thermal
capacity, it is more important to know the peak demand than to know the
amount of electrical energy demanded, since the peak demand often sets the
capacity expansion goal.
For systems with large amounts of hydroelectric capacity, it may be more
important to know energy demand because these systems may have energy
limitations. Knowledge of the peak demand is also important for planning the
type of generating capacity that should be built, when it should be scheduled
for maintenance, and how much reserve will be needed (both spinning and
standby). Milkias B. 8/11/2020 58
 …
Example: Peak Power Requirements of the Ethiopian grid electricity
supply grows from in 2006 to 4,800MW in 2016

Milkias B. 8/11/2020 59
Estimation of Generation requirement

Again usually energy demand data is available sectorially in the form of sales
(belled energy) data. This has to be changed to energy requirement by assuming
a certain percentage of loss.
Generation= Sales + losses
Losses=x%Generation
Sales
Generation =
1 − x%
Peak load Forecast
The system peak load is derived by the assumption of the future trend
of the load factor
AnnualEnerhy
PeakLoad =
8760 * Loadfactor
Scenario technique
In the scenario technique several different forecasts are made for electric
energy demand. These forecasts are usually based on assumptions concerning
the independent variables. This method helps to study the influence of
different factors on energy demand. Milkias B. 8/11/2020 60
Other Load Forecasting Techniques

▪ The techniques used and implemented to create STLF

are:
▪ 1. Linear Regression
▪ 2. Stochastic time series
▪ 3. General Exponential Smoothing
▪ 4. State Space Method
▪ 5. Knowledge – based expert approach
▪ 6. Artificial Neural Network (ANN)

Milkias B. 8/11/2020 61
 LOAD FORECASTING THROUGH ARTIFICIAL
NEURAL NETWORK
▪ Load forecasting is a central integral process in the planning and operation of
electric utilities.
▪ Load forecasting has become in recent years one of the major areas of
research in electrical engineering.
▪ The main problem for the planning is the determination of load demand in
the future. Because electrical energy cannot be stored appropriately, correct
load
forecasting is very essential for the correct investments.
▪ Effective forecasting, however, is difficult in view of the complicated effects
on
load by a variety of factors such as temperature, humidity.
▪ A three-layered feed forward neural network can be trained by the Levenberg-
Marquardt algorithm and a radial basis function using matlab programming
and matlab tool-box.
▪ The Proposed neural network based model is used for forecasting next-week
electricity prices. We
Milkias B. 8/11/2020 62
 Neural network approach
▪ Neural networks are highly interconnected simple processing units designed in
a way to model how the human brain performs a particular task.
▪ Each of those units, also called neurons, forms a weighted sum of its inputs, to
which a constant term called bias is added.
▪ Fig. I shows the internal structure of a neuron.
▪ Multilayer perceptron are the best known and most widely used kind of neural
network.
▪ Networks with interconnections that do not form any loops are called feed
forward.
▪ Recurrent or non-feed forward networks in which there are one or more loops
of interconnections are used for some kinds of applications
▪ The units are organized in a way that defines the network architecture.
▪ In feed forward networks, units are often arranged in layers: an input layer,
one or more hidden layers and an output layer.
▪ The units in each layer may share the same inputs, but are not connected to
each other.
▪ Typically, the units in the input layer serve only for transferring the
input pattern to the rest of the network, without any processing.
Milkias B. 8/11/2020 63
...
The information is processed by the units in the
hidden and output layers.

Milkias B. 8/11/2020 64
 METHODS OF LOAD FORECASTING
▪ Electric load forecasting is challenging because of the different characteristics of
the electric loads associated with them.
▪ There are various methods used to forecast electric load. They are broadly divided
into 2 main groups. They are: statistical methods which involve a lot of
mathematical analysis and expert systems (which is the use of computer for load
forecasting).
▪ The methods mostly used are discussed below:
(a) Simple Statistical Method
▪ Statistical models usually require a mathematical model that represents load as
functions of different factors such as time, weather, and customer class.
▪ The important categories of such models are additive and multiplicative models
(b)Expert Systems
▪ Expert systems incorporate 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. This means this program can reason, explain and have
its knowledge base expanded as new information becomes available to it.
▪ The load-forecast model can be built using the knowledge about the load
forecast domain from an expert in the field
Milkias B. 8/11/2020 65
 ...
▪ Load forecasts are extremely important for energy suppliers, ISOs, financial
institutions, and other participants in electric energy generation, transmission,
distribution, and markets.
▪ Load forecasting is however a difficult task because the load series is complex
and exhibits several levels of seasonality systems.
▪ Daily maximum load forecasting is used for the applications like the unit
commitment, security analysis of the system and the economical scheduling of
the outages and fuel supply.
▪ Artificial Neural Network (ANN) is mostly used for the prediction of the load.
The reason is, ANN methodology solves the complex relationships between the
independent and dependent variables by a mathematical mapping algorithm.
▪ Short-term load forecasting (one hour to one week ahead) plays a key role in
economic and secure system operation. Short term load forecasting displays a
great ability for economic and secure operation of power.
▪ In the rapidly growing power markets like India with the limited generation
capacity, this can be the powerful tool for the demand side management. Efforts
are made in this work to develop, train and test an artificial neural network model
which can forecast the peak load to a reasonable accuracy of a smaller area with
data knowledge.

Milkias B. 8/11/2020 66
Artificial Neutral Network (ANN).

• ANN is originally developed to mimic basic biological neural systems.

• An Artificial Neural Network (ANN) is an information processing
paradigm that is inspired by the way biological nervous systems, such as the
brain, process information.
• The key element of this paradigm is the novel structure of the information
processing system. It is composed of a large number of highly
interconnected processing elements (neurons) working in unison to solve
specific problems.
• ANNs, like people, learn by example. Neural networks are essentially
nonlinear circuits that have the demonstrated capability to do non-linear
curve fitting.

Milkias B. 8/11/2020 67
 ANN forecasting using model
▪ The developed forecasting model, which considers
temperature and humidity as input data, is shown in Fig. 1.
▪ Since the load demand does not depend only on temperature
and humidity and in order to account for other factors, the
input data (temperature and humidity) are fed to the neural
network with historical load data for training and comparing for
future load forecasting.

Milkias B. 8/11/2020 68
 ...
▪ It is a simplified model of the central nervous system and is form of
artificial intelligence.
▪ Neural network is an attempt at creating machines that work similar
way to the human brain by building these machines using components
that behaves like biological neurons.
▪ Neural networks are interconnected neural computing elements that
have ability to respond to input stimuli and to learn, to adapt to the
environment.
▪ This project developed a very simple model for predicting the electric
load from 2013 to2027 in using Artificial Neural Network.
▪ The table below indicates the monthly electric energy consumption
from January 2008 to February 2013.

Milkias B. 8/11/2020 69
The Simulink Model was extracted from the net fitting toolbox is shown below.

Fig. Simulink Model

Milkias B. 8/11/2020 70
Table 1: Table of the monthly electric energy consumption

Milkias B. 8/11/2020 71
FORECASTING RESULTS

Milkias B. 8/11/2020 72
…

Milkias B. 8/11/2020 73
 ...
▪ The graph in the figure 1 showed the performance of the network during
the training process (set by the number of epochs used).
▪ During the training process, the data was divided into three, train data,
test data and validation data.
▪ For this model, the trained data was 70% of the total data and others
were 15% each. These sets of data were used to adjust the parameters of
the network for the best performance during the process.
▪ Training stopped when a set validation criteria was met, i.e. where all
data sets were at their lowest, which in this case occurred at the 10th
epoch.
▪ The graph in figure 2 showed the forecasted result from 2013 to 2027,
there was a gradual rise in power consumption based on existing power
consumption pattern, it was expected that by 2027 the load consumption
will rise to1.94 x 105 kwh.
▪ Also table 2 indicated the monthly loads consumption forecasted from
2013 to 2027 in kwh.
Milkias B. 8/11/2020 74
Training and Testing with ANN
▪ The whole data set was divided into two sets: Training set and
Test Set. Training set consists of 80% of whole data and Test set
contains the rest data. The training set was used to make a model
which, therefore, predicts the load in the future. The model is
made by a MATLAB app Neural Net Fitting. The training set has
got inputs which are as follows:
▪ 1. Temperature (in oC)
▪ 2. Humidity (in %)
▪ 3. Pressure (in mBar)
▪ 4. Time (in hours)
▪ 5. Global Horizontal (in W/m2)
▪ 6. Previous Day Same Hour Load (in kW)
▪ 7. Previous Week Same Day Same Hour Load (in kW)

Milkias B. 8/11/2020 75
 ...

Fig 2: Forecasted load consumption from 2013 to 2027.

Milkias B. 8/11/2020 76
Table 2: Complete loads forecast result from 2013 to 2027 in KWhr

Milkias B. 8/11/2020 77
Milkias B. 8/11/2020 78
 ? ?
? ?
? ?
? Any Questions ?
and
Discussions ?
? ? ? ?
? Milkias B.
?
8/11/2020
79