Short and Mid-Term Wind PowerPlants
ForecastingWith ANN
Javad Mahmoudi
MSc student of Electrical Engineering
of Sharif University
Tehran, Iran
E-mail: Jmahmoudi66@gmail.com
Majid Jamil
Material and Energy Research
Center(MERC), Department of Energy
Karaj, Iran
E-mail: m-jamil@merc.ac.ir
Hossein Balaghi
MSc student of Mechanical
Engineering of Sharif University
Tehran, Iran
E-mail: Balaghi.hosein@gmail.com
Abstract—In recent years, wind energy has a remarkable growth
in the world, but one of the important problems of power
generated from wind is its uncertainty and corresponding power.
For solving this problem, some approaches have been presented.
Recently, the Artificial Neural Networks (ANN) as a heuristic
method has more applications for this propose. In this paper,
short-term (1 hour) and mid-term (24 hours) power forecasting
are presented for a sample wind power plant by multilayer ANN.
The needed inputs data are temperature and wind speed for
forecastingthe power. A case study has presented.
Keywords-wind power forecasting; artificial neural network;
power prediction
I. INTRODUCTION
Due to rising consumption of electricity, reduction of fossil
fuel and its high pollution in world the orientation of
Generation Companies (GenCo) of electricity energy have
accelerated to renewable energies more than ever. Wind energy
is important and available in most regions but due
touncertainty wind speed and direction,the prediction has a
significant role in power generation[1].
The basic problem of connection wind power to grid utility
is forecastingin competitive market.Thereforpower forecasting
tools will be needed [2],[3]. Precise prediction of power helps
dispatch easily wind power and determines that how much
reserve system should be consider for supporting of wind
power units [4].For the purpose of maintenance, the forecasting
is important for electricity industry, e.g. when one turbine must
be disconnected or repaired for improvement of system
performance. The important of wind forecasting is clear for
dispatchability and cost of producing electricity from wind.
Current and future challenges to forecasting wind power are
integrating and automating regional forecasts with electricity
scheduling systems and incorporating climate change impacts
on wind projects[5].As the amount of wind energy requiring
integration into the grid increase, short-Term forecasting
became more important to both wind farm owners and the
transmission as well as distribution operators[6].
II. HEURISTIC METHODES
Due to high performance of heuristic methods for
algorithms identification, they can be used to predict output
power of wind turbines. Artificial neural networks classify
information content as significant by analysis of input and
output (I/O) data. Outputs are forecasted after training system
with I/O data by new inputs [7]. In this paper, a new method is
presented for one hour and one day ahead forecasting of wind
turbine output power.
III. ARTIFICIAL NEURAL NETWORK(ANN)
The ANN is commonly used between heuristic methods.
The first neural cell was used by Mac Lorch and Pittz in 1943.
This method can be trained and extract nonlinear complicated
relationbetween input and output. Since 1990s, ANN is entered
in engineering science vastly [8].
These methods have high performance in estimation and
approximation of real engineering systems [9].
Figure 1 shows structure of a neural processing unit and
relation units.
All inputs (X
N
) are weighted with W
N
at first part of cell.
Combination Function (+) sumsall weighted inputs at every
nodes and adds a threshold value or bias () for changing its
status. The bias increase and decrease sum of weighted inputs
Figure 1. Structure of a neural processing unit and parameteres[8]
Transfer
Function
2012 Second Iranian Conference on Renewable Energy and Distributed Generation
978-1-4673-0665-2/12/$31.00 ©2012 IEEE 167