Abstract- Paper gives the compressed chronological work of scholars and comparison of classical and modern techniques to short term forecast the electrical load.
This paper presents 15 approaches of load forecasting, their brief introduction and literature review.
Keywords- Load Forecasting, Classical Approach, Heuristic Approach, Fuzzy Logic, AI, ARMA,
I. Introduction
Earlier the utilization of electrical energy was simple and straight forward; generation of electricity & it's consumption was a linear process but now a day's electrical power system is one of the complex network in world, increase in demand with diversification, connecting various regional grids and set up of a centralized control like National Load Dispatch centre (NLDC) are the mile stones in this up gradation [1]. Electrical load forecast is simply means how much amount of electrical energy requires in future and this will play in important role in deciding the capacities and size of generation, transmission and distribution system [13]. Load forecasting is vitally important for the electric industry in the deregulated economy. It has many applications including energy purchasing and generation, load switching, contract evaluation, and infrastructure development [ ].
Mostly in developing countries where the resources are limited and coordinate planning has been a crucial task load forecasting is a boon [3]. Load forecasting and its techniques are no longer stranger term in India at present era although it had been the most neglected part [7] since last few years but in few years the privatization has been taking part very rapidly in generation and distribution sector and with the set up of agencies like PXIL[ ] who are solely responsible for power exchange and trade in India; electrical load forecasting got a new horizon and most of the utilities have been using online software for load prediction carrying SCADA[10], State estimation [ ], Phasor Measurement Unit(PMU)[ ] and other similar connecting and control network for power flow and allied activities.
The aim of this paper is to review and classify short term electric load forecasting methods. [8, 25, 26,] have discussed the review of STLF. The earliest survey for load forecasting was carried out by Matthewman and Nicholson (1968) [25]. After Moghram and Rahman (1989) [26] who has done review for load foresting techniques; many new techniques has introduced. Hasham and Mohammad (2002) [8] has given an excellent survey and comparison of different load forecasting methods. Shu Fan et al(2011)[14] gave the comparative study on load forecasting techniques for various loads distributed geographically. Load forecasting for large geographical area with multi area forecasting is done by Fan S. et al (2008) [9]; provide a clear guideline for load forecast of different category under one roof.
Consideration attention has been paid by researcher to survey the electric load forecasting technique [28, 29, 30].
The load forecasting is totally based on time and hence classified under three categories [1]:
Short term load forecasting
Medium term load forecasting
Long term load forecasting
Short term load forecasting covers the time interval of 1 hour to 1 week and is mainly uses to losses reduction, Voltage regulation, power system operation and optimization. Medium term load forecasting cover the time interval of 1week to 1 year, it predicts the necessary electrical power required to purchase or to sell in case of access along with fuel or resources required for electricity generation [27]. The period of 1 year to several years is considered for long term electric load forecasting, It is utilizes the infrastructure to be build for generation, transmission and distribution, resources required as fuel, land, conductors, other auxiliaries and man power for their commissioning and operation.
For instant power saving and advantage short term and very short term load forecasting (time shorter than one day) are more considering by utilities in distribution network, Organizations which are involve in power generation give concentration on medium term load forecasting for management of fuel and man power while Statuary agencies like Electricity Regulating Commission give more emphasis on Long term forecasting for future perspectives.
The paper is organized as follows: Section 2 describes the typical load model as well as the typical load affecting factors. Section 3 discusses the key components of an artificial neural network. Section 4 focuses on the development of the forecasting model. In section 5, the data obtained from the Cape Town Control Centre is used to test the model that was developed. Section 6 presents the conclusions.
II. Load forecasting techniques and comparision
An electrical load forecasting method or model contains mathematical representation and description of complex elements of real situations [ ] .
The classification of load forecasting techniques is done by different scholars as per their conventions few of among them is discussed here.
These techniques are classified as classical and modern.
The above classification may categories in the following methods [2] [3] [7] [8] [27]:
Multiple Regression Method
Exponential Smoothing
Iterative Reweighted Least Square
Adoptive Load Forecasting
Stochastic Time Series
Armax Model
Similar Day Look Up Approach
Statically Robust Method
Delpha Technique
End Use Analysis
Fuzzy Logic
Artificial Neural Network
Support Vector Machine
Expert System
Econometric Method
Multiple regression method
Multiple regression (Matthewman 1968) [25] uses dependent variable to determine independent variable through modeling.
The load should found in terms of independent as well as dependent variables, the model uses the form as
Where
y is the load,
x1, x2 & xk are the affecting factors
ï¢ï€°ï€¬ï€ ï¢ï€±ï€¬ï€ ï€¦ï€ ï¢ï«ï€ are the regression parameters with respect to x1, x2, & xk
ï¥ï€ is an error term
Parameter ï¢ï©ï€ are unknown & have to known on observation of y & x
This method is very sensitive to temperature variation and a small change may give a considerable change in forecast [33]. A change in temperature in directly responsible for load variation; the calculation and simulation for this in bangladesh power system in done by Nahid-Al-Masood et al(2010)[18] with linear regression model.
Exponential Smoothing
This method uses data of the energy consumption in history and based on time series (Christiaanse 1971). Smoothing method smoothed the data for estimates. Following types of exponential smoothing is using presently [34]:
First order exponential smoothing;
Second order exponential smoothing;
Higher order exponential smoothing;
Holt-Winters mechanism
Iterative reweighted least square
Method uses an operator to control one variable at a time (Mbamalu and El-Hawary 1992) [8]
The starting point finds through operator. The method takes autocorrelation function as well as partial autocorrelation function to find a model of load dynamics. A three way optimal model is identified; weighting function, tuning constant and weighted sum of the squared residuals.
Consider a linear equation for parameter estimation:
Where
Y is vector of observations of n X 1
X is a coefficient of n X p
ï¢ï€ ï€ is unknown parameter of p X 1
ï¥ï€ ï€ is random error of m X 1
Mbamalu and El-Hawary (1993) proposed an interactive approach employing least-squares and the IRLS procedure for estimating the parameters of a seasonal multiplicative autoregressive model.
Adoptive load forecasting
Adoptive load forecasting is based on continuous correction of data. It can be use as online software system for utilities to control the flow of power which uses the Kalman filter theory with current prediction error and the current weather data acquisition programs to estimate the next state vector [8].
Stochastic time series
Time series method has been most popular method although it has several draw backs such as complex to use, require more time and historical data for prediction but in today's most complex system and system of fast development in context of energy generation and demand method has difficulty to predict however it has been using for STLF (Hegan 1987) [36][8] . The remaining models of time series uses are:
Autoregressive (AR) model
Autoregressive moving-average (ARMA) model
Autoregressive integrated moving-average (ARIMA) model
Similar day look up approach
Similar day look up approach uses the historical data of few years back which has similar weather conditions, similar days in week, similar regional activities. Now a day's forecast through similar day look up approach uses combination of several days through linearly or through regression. The technique has its own limitations as the demographic situation for similar day may get change and it will deflect our forecast from accuracy.
Fuzzy logic
The fuzzy logic method for load forecasting has accuracy and less time requirement for computation and there is no need of mathematical models for mapping between input & output. For precise output & unknown dynamic system; which is possible in fuzzy logic through centroid defuzzification [].
Fuzzy logic has an excellent capacity to draw similarities from same data [40].
Liu et al. (1996) [41] saw that fuzzy system is capable in drawing similarities from a large amount of data which may define as:
Where First order difference
Second order differences
There have been two stage always have been include; training and online forecast work [8]. During training session measured data of previous load is used to trained the system with selection of variables in linguistic form where the fuzzyfication is processed are forecast is done through approximation and then efuzzyfication is carried out to show output. The block diagram shows its process system [13]:
Process flow of data through FIS
The input variables are all those which ultimately affect electricity consumption are as day's maximum temperature, day's minimum temperature, working day, holidays, day's capacity etc.
Artificial Neural Network
In neural network, the basic element is neuron. The neurons get information from a source and combine and perform operation with them and produce result [8]. Artificial neural network were starting development in mid 1980s; still these are in very early stage although we find an extraordinary perforation from the system based on pattern recognition observed from past event and estimations the values for prediction in future. ANN is capable to approximate any faction numerically for required accuracy and is data driven method which provide facility to user that tentative model. Liu et al. (1996) uses a fully connected feed forward type neural network[41 ] where Damborg et al (1990)[40] wrote neural network offers potential to overcome reliance on functional form of a load forecasting model. Daniel Ortiz-Arroyo et al.(2005)[ ] gives accurate load forecasting with ANN with analyzing the problem and choosing appropriate domain. K. LIU et al (1996) [41 ] gives the comparative study on methods of load forecasting specialy ANN. Hesham K. Alfares et al (2002)[8] gives an excellent comparative study on ANN models and previous work on it. D.C. Park et al (1991) [39] gives ANN model for electrical load forecasting. Paras Mandal et al
(2006) [37] gave a several hour ahead load forecasting model using ANN. Y. Rui et al gave a review on comparative models on ANN. Guoqiang Zhang et al(1998)[38 ] shown a state of art load forecasting using ANN. G.A. Adepoju et al (2007)[ ] gave application of ANN for load forecasting in Nigerian power system. Sanjib Mishra and SK Patra(2008) [17 ] forecast electrical load using neural network and hybrid it with artificial immune system. Load forecasting with ANN is more closer to real load but at peaks the errors are more; at these peaks the reduction in error is possible using rough set conception[36]
Knowledge based expert system
Expert systems are the result of advancement in Artificial Intelligence in last two decades. It is software program which acts as expert system. The program has the ability to reason, explain and expand new information. The model for load forecasting is created using knowledge from experts. Above is called acquisition module component of expert system.
Rahman et al (1991)[ 16] gives the rule based algorithm which consists of functions developed for electrical load forecasting based on relation between weather and prevailing daily and hourly load shapes; this happens in rule base form of IF-THEN. The work is performed offline and dependent on operator experience and observation. A decision making model for combination of short term load forecasting is given by Kang Chongqing et al (2012) [15] which is a key feature of expert system.
Econometric method
The method is a combination of trend analysis and end use analysis which uses mathematics, economics & statistics for electrical load forecasting. Econometric method makes changes possible in relation of input and output. It uses the complex equations for creating relation between input and output for load forecasting the method has a great advantage that it can predict the load for different segments like residential, commercial or industrial sector which ultimately reduce the error in overall forecasting. Present agencies of government mostly believe and apply this method for load prediction; fig. gives brief information about residential segment with influencing factor [6]:
Residential forecasts are influenced by six factors: per capita income, population, residential electric prices, residential gas prices, household types, and efficiencies. The uncertainty of load forecasting concerning with market price and power system reliability is given by Kang Chongqing et al (2006)[10]
IV. Conclusion
This paper present review of the recent development in the area of Electrical load forecasting. Emphasis has been given to categorizing various short term load forecasting methods which is reported in the literature. Paper also presented salient feature of the various short term electrical load forecasting methods. This paper will serve as a valuable resource to any future worker in this important area of research. The uncovered subject material on electrical load forecasting is available in books [60, 61, 62, 63], tutorials [ ], review [64, 65, 66, 67], and state of the art lecture [ ].
