Volume 4, Issue 8, August – 2019 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Electrical Load Forecasting in Power System

1 2
Olajuyin E. A. Olubakinde Eniola
Elect. & Inform. Engineering, Electrical and Electronics Engineering
Covenant University Ota, Ogun State Federal Polytechnic, Ile-oluji, Ondo State

Abstract:- Load forecasting is normally employ to operation of any utility. Forecasting in the countries that are
forecast and predict the rise in power demand of an area just developing is extremely complex due to the large
in other to prepare for electricity generation to meet the differences in the modes and the rate of energy required from
costumers demand as the population is growing. Electric one location and increase growth rate of load demand.
load Forecasting can be used in selling, planning and Electric load forecasting is extremely important for energy
buying of energy in power systems. It is very useful from suppliers, financial institutions, and other participants in
generation to distribution of electrical energy. electric energy generation, transmission and distribution
Forecasting in power system can be widely divided into (Feinberg and Dora ,2002). Electric load Forecasting is used
three classes: Short-term electric load forecasting is selling, planning and buying of electric energy in power
generally one hour to one week, medium electric system. It is very pertinent in the power industry and it has a
forecasting is normally from a week (7days) to a year, and major economic advantage as it helps the power generating
while long-term electric load forecasting which are may and Distribution Company to estimate the amount of power
be from a year and above. Accurate load forecasting can needed to be generated and supplied to a particular
help in developing a plan, particularly for developing geographical area since electric energy cannot be efficiently
countries in which energy demand in such countries is stored. It is a relevant tool in taking of decision and planning
irregular in nature as result of rapid growth rate and by the utility. It is also used in the total planning of the
increase in the rate of rural–urban drift. The different system.
methods of load forecasting were studied such as expert
system, fuzzy logic, regression methods and artificial Accurate load forecast can help in developing
neural network (ANN).Some benefits of electric load development plan, energy supply policy, and especially for
forecasting were discussed. developing nations in which the energy demand is irregular
in nature because of the increase in growth rate and rural–
Keywords:- Regression, Forecasting, Load, Demand, Power, urban migration.
Electric, Generation e.t.c.
Other merits of electric load forecasting are:
I. INTRODUCTION  Good investment and network preparation
 enhanced and heuristic running of risk
In developing the economy of a country today, there is  decrease in expenditure
need to engage in electric load forecasting and accurate
prediction of electric future load demands will help decide Forecasting can be classified according to both the
and calculate approximately the amount of energy needed so predicted quantity (peak load, integral load, hourly load) and
as to get ready for power generation (Idowu ,2009). Demand the prediction time. Based on the prediction time, it can be
energy forecasting is a process of calculating the electric categorized into three classes. They are:
energy necessary by a particular area by using previous
records of energy usage in that location and some other  Short term load forecasting: This is generally one hour to
relevant conditions such as temperature, weather forecast and a week. It helps to guesstimate the load flows and to make
population, etc. The issue in demand forecasting is to obtain solid decisions that can avert overloading. It is employing
what will be needed in future demand and by mainly in controlling, in power scheduling, input to load flow and
extrapolating past consumption and other factors which can also contingency analysis.
affect the amount of energy that is been used in an area at  Midterm load forecasting which is normally from a week
that specific period (Adepoju et al ,2007). to one year; and also
 Electric long term load forecasting is from one year and
The verity that electric power cannot be stored to be above. Long term electric load forecasting is used to
used at the later time with the available technology and verify the capacity of power system.
facility, this made it necessary to have the understanding of
the way electric load demand behaves in the future. Electric Some of the factors that influence forecasting include:
utility planning begins with energy forecasting because of weather data, time factor, socio-economic settings (festivities
necessitate for new utility stations. Precise models for electric and working days) and special periods such as Christmas or
load forecasting are indispensable to the planning and Ramadan, past demand load data, and the type of location

IJISRT19AUG775 www.ijisrt.com 845

Volume 4, Issue 8, August – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
(Ibitoye and Adenikinju, 2006). All these factors stated above method employed steady state component or average
will decide the amount of energy that would be need by an component. However the calculation of the load depends
area at any specific period. Forecasting is a major ingredient upon the weighted grouping of the components which varies
of resolution making. Its function is to lessen the threat in dynamically (Amit and Satish, 2008).
decision making and lessen unforeseen expenditure.
Information obtained from short-term electric load  Support Vector Machines
forecasting is important to power supply system as operations Support vector machine (SVM) is a mew powerful tool
in conditions of short-term part maintenance service, daily, to solve regression and classification problems. They execute
hourly and weekly load scheduling of generating station, non linear mapping of the data into a elevated dimensional
secure operation of power systems and economic space. Then support vector machines uses easy linear
(Gangshem,2000). functions to form verdict boundaries in the new space.
Support Vector Regression is an optimization problem. The
 Electric Load Forecasting method support vector method compares favorably with the
The forecast of electric load is demanding because of regressive method (Weichang, 2009). The predicament of
the different distinctiveness of the electric loads related to choosing building is changed by the setback of choosing a
them. There are different methods employing in electric fitting kernel for the support vector machine. It is employed
forecasting of load. They are widely classified into two major in short term forecasting of loads
categories: Statistical methods and expert systems.
 Artificial Neural Network
These various methods that can be used are explained This is a trouble-free model of the middle nervous
below: system and it is part of artificial intelligence. It is an attempt
to form machines that function in related way to brain of
 Simple Statistical Method human beings by constructing these machines employing
The statistical methods normally need a mathematical apparatus that perform like biological neurons. Neural
model that depicts load as functions of various factors such network is interconnected neural computing basics that have
as consumer class, weather and time. The significant the capacity to reply to input stimuli, to adapt and to learn the
categories of such models are multiplicative models and environment. It can be educated to solve different problems
additive (George and Burton, 1979). The forecast load is a which are not easy to solve with human brain or conventional
computation of a number of components and result of a many computers. After a neural network has been trained, it has the
factors. The advantages of this method include its simplicity ability to make predictions.
as it involves minimum number of variables. It is also the
easiest to compute but has the disadvantage of not being Parameter Definition
accurate. It can be used for any of the types of forecasting. LW Layer Weight
IW Input Weight
 Regression Methods F Network function
Regression methods are normally employed to model P Input
the relationship of energy consumption and other factors such B Network bias
as type of day, consumer type and weather (Hafiz et al Table 1:- Representation of variables of neural network
,2002). Regression study performs linear regression using the equation
least squares method to fit a line through a set of remarks. It
is very common in short term forecasting of load.  Training, Network Parameters and Error Correction
Back propagation is the most universal form of
 Expert Systems learning. The weights are distorted based on their correction
Expert systems integrate rules and measures employed term and previous value. It is a means of error correction that
by human experts in the field of interest into software which can be employed to adjust the weights of the connections in
robotically make forecasts without assistance of human hidden layers. The learning law is the way in which the
beings. The program has its knowledge base extended as new correction condition is formulated. It is basically a steepest
information available to it, reason, perform and dive process. For the gradient dive algorithm, a step size
explain.(Amit and Satish, 2008). (1.000 step size was used), that is known as learning rate in
ANN writing must be indicated. The learning rate is
 Fuzzy Logic important for back propagation in learning algorithm since it
Fuzzy logic is a overview of the usual Boolean logic will decide the size of weight alteration, (Uduehi, 2009).
employed in digital circuit design. in fuzzy logic, inputs have
connected with it assured qualitative ranges. It allows one to It is very known that the steepest dive endures the
logically assume outputs from fuzzy inputs. It performs on challenges of lack of robustness inefficiency and slow
the foundation of mapping input to output. Moreover, this convergence. Moreover, it is sensitive to the learning rate.

IJISRT19AUG775 www.ijisrt.com 846

Volume 4, Issue 8, August – 2019 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Smaller learning rates tend to sluggish the learning procedure  Run The Network
while larger learning rates may cause network oscillation in Entering new input into the network and gather usable
the weight space. result.

In back propagation, to discover the error value for a II. CONCLUSIONS

given node h in a hidden layer, you need to get the weighted
sum of the errors of all nodes linked from node h i.e., of all Electric utility planning begins with electric forecasting
nodes that have an incoming association from node h of load of require by the consumers the utility will be
h = w11 + w22 + w33 + … + wnn (3.2) supplying. Precise models for electric power forecasting of
load are fundamental to the planning and operation of any
Where n is the error between the actual value the utility. Different methods discussed above have been applied
output from node h. to forecasting of load. The merits and demerits of these
wn is the weight of the connection to node n methods are discussed .All the methods discussed above can
h is the error of the hidden layer node (Jaap,1999) be applied in forecasting of electric loads in generation,
transmission and distribution of power system.
So the Weight Change = some constants  input activation x
error REFERENCES

Output node, the error is given as: [1]. Adepoju, G.A, Ogunjuyigbe, S.O.A. and Alawode,
Error = (target activation - output activation)  (1 - output K.O. 2007. “Application of Neural Network to Load
activation) x output activation Forecasting in Nigeria Electric Power System”. Pacific
Journal of Science and Technology. 8(1):68-72
The hidden node, the error is given as [2]. Amit, J, and Satish, B. 2008. “A Hybrid approach for
Error = weighted sum of to-node errors  (1 - hidden short term load forecasting using SVM and Time series
activation) x hidden activation (Uduehi, 2009). Technique” 15th National Power Systems Conference.
Bombay: 262-268
 How to use Artificial Neural Network in Forecasting of [3]. Feinberg, E.A., Dora Genethliou. 2002. Load
loads Forecasting; Applied Mathematics for Power Systems;
The following steps must be taken when using artificial 269-285.
neural network: [4]. GaangShen, EE653 Project 2000 #1, P. 1 – 2
[5]. George, O. and Burton, E. 1979. “A review of load
 Idea forecasting methods” Card Series, 84-5
A better understanding of what the network is expected [6]. Hafiz, H.A.S., Ahmed, H., El-Debeiky,S. 2002.
to give must be well cleared. “Electric load forecast for developing countries” IEEE
MELECON 429 – 440
 Information [7]. Ibitoye, F.I. and Adenikinju. A. 2006. “Future Demand
for Electricity in Nigeria” Applied Energy 84 (2007)
The information required for the predictions must be
492 – 504.
decided.
[8]. Idowu,O.M. 2009. ”Long Term Electric Load
forecasting using regression analysis”, B.sc. Project
 Data Collection
University of Lagos.1-8.
The information required must be obtained from the
[9]. Jaap,M.1999.” Learning Neural Networks: Perceptron
available source.
and Backpropagation”, p 23 - 34
[10]. http://www.neuromod.org/courses/connectionism/backp
 Building a Network ropagation
The number of input and output neurons must be [11]. Uduehi,O.2009.”Load Forecasting Using Artificial
specified when building a neural network. Neural Networks,”M.Sc project University of Lagos.
[12]. Wei-Chang, 2009 “Electric Load Forecasting by
 Training the Network Support Vector Model” Applied Mathematical
The actual load generated is used as input load to train Modelling 33; 2444-2454.
the network which takes in each input and makes a guess as
to the output.

 Testing the Network

It is important to test a network on data it has never
seen.

IJISRT19AUG775 www.ijisrt.com 847