Wind energy forecasting using radial basis function neural networks

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 04 Issue: 12 | Dec-2015, Available @ http://www.ijret.org 274
WIND ENERGY FORECASTING USING RADIAL BASIS FUNCTION
NEURAL NETWORKS
P. Badari Narayana1
, R. Manjunatha2
, K. Hemachandra Reddy3
1
Director, Green Life Energy Solutions, Secunderabad, T.G, India.
E-mail: bnarayan@gles.in
2
Asst. Executive Engineer, Department of Irrigation & CAD, GBC Division, Guntakal, A.P, India.
E-mail: manju.biodsl@gmail.com
3
Professor, Department of Mechanical Engineering, Registrar, JNTUA, Anantapur, A.P, India.
E-mail: konireddy@gmail.com
Abstract
Wind power forecast is essential for a wind farm developer for comprehensive assessment of wind potential at a particular site or
topographical location. Wind energy potential at any given location is a non –linear function of mean average wind speed,
vertical wind profile, energy pattern factor, peak wind speed, prevailing wind direction, lull hours, air density and a few other
parameters. Wind energy pattern data of various locationsis collected from a published resource data book by Centre for Wind
Energy Technology, India.Modeling of wind energy forecasting problem consists of data collection, input-output selection,
mappingand simulation. In this work, artificial neural networks technique is adopted to deal with the wind energy forecasting
problem.After normalization, neural network will be run with training dataset.Radial Basis function based Neural Networks is a
feed-forward algorithm of artificial neural networks that offers supervised learning.It establishes local mapping with two fold
learning quickly.Wind power densities predicted for new locationsare in agreement with the measured values atthewind
monitoring stations.MAPE was found out to be less than 10% for all the values of Wind Power Density predictions at new
topographical locations and R2
is found to be nearer to unity.WPD values are multiplied by wind availability hours (generation
hours) in that particular location to give number of energy units at the turbine output. These values are compared to the output of
the wind turbine model installed in the same region, so as to assess the number of units generated by that particular wind turbine
in the respective locations.This kind of assessment is useful for wind energy projects during feasibility studies. With this work, it is
established that radial basis function neural netscan be used as a diagnostic tool for function approximation problemsconnected
towind energy resourcemodeling& forecast.
Keywords: Wind power density, wind energy, forecast, modeling, air density, peak wind speed, radial basis function,
neural network, CoD, MAPE
--------------------------------------------------------------------***----------------------------------------------------------------------
I. INTRODUCTION
Wind resource assessment is a preliminary requirement for a
wind farm developer.Identification of potential windy sites
is made within a fairly large region of several kM2
by
analyzing data comingfrom meteorological stations.
Climatological data obtained from these stations along with
topographical maps and satellite images is combined for this
purpose. During this stage, wind resource assessment
methodology is termed as Prospecting. The next stage is
Validation Process.It involves more detailed investigation
such as wind measurements and data analysis. The final and
necessary step is Micro survey and Micro siting. The main
aim of this step is to quantify the small scale variability of
the wind resource over the region of interest, i.e. 10 KM
radius from wind monitoring station taken vertically and
horizontally.At the end, micro siting is carried out to
position the wind turbine in a given area of land. Efficient
modeling of wind resource data post the Validation Process
maximizes the comprehensive energy output of the wind
farm [20].Wind velocity and direction measurements are
obtainedas every 1 second or 2second datasets and
areadjusted as 10 minute average values. These valueswill
be mentioned withminimum, maximum and standard
deviations. This data will be used by Wind generator
manufacturers to validate the power curve of their turbine
models. Actual energy generated by the wind turbine will
surely show some deviation from the manufacturer’s power
curve due to many reasons. Probabilistic determination of
wind speed data above 25 m level, standard deviation
methodology, error in air density extrapolation across the
height and measurement slags are some of the reasons for
these deviations [16]. Also, wind power economics is
greatly influenced by mean annual energy produced per
turbine which in turn depends on wind resource variation
across different topographical locations vis-à-vis seasonal
variations. Hence wind resource assessment and forecasting
is a prima facie activity for the development of a wind
farm.Selection of a suitable location for a wind plant
primarily depends on wind resource estimation at a
particular topography based on meteorological, atmospheric
and geographical parameters. Wind resource assessment of
any place is characterized by wind speed, direction, vertical
wind profile, turbulent intensity, air density and

_______________________________________________________________________________________
topography.Terrain roughness, tunnel formations, nocturnal
variation, diurnal variation, monsoon variation and many
other climatological factors also contribute for wind energy
assessment.In this paper, artificial neural network technique
is adapted to model wind power density prediction problem
with influencing factors being considered from wind
resource mapping data of monitoring stations.
II. WIND RESOURCE PARAMETERS
The basic data required to estimate wind energy potential for
any topographical region consists of wind velocity and wind
power density values. They are determined at the hub height
of the turbine. The resolution available should be 20mtrs or
50 mtrs. The wind resource parameters measured during the
wind resource assessment programmes are as briefed below:
A) Mean annual wind speed: This is the mean value of
measured wind speeds across one year data points.
Mean annual value wind speed is defined asu.
u = 1 ∑ where ui = individual wind speed
n is sample size of wind patterns.
B) Wind power density
The wind power at any location is given by
= /2
-here p is the air density. So the available mean wind
power at any location is given by
E(U)=ρE( )/2
C) Power Law Index:
Power Law wind profile is expressed as:
U2 = U1 (Z2/Z1)α
U2= wind speed taken at level Z2
U1 = wind speed taken at level Z1
Z1 = Reference height
Z2 = Computed height
-where ‘α’ is power law exponent. The power law index
is a function of surface roughness and stability.
Empirical studies have found that a value of 0.14 best
fits to most of the sites. This method is always referred
to as one-seventh power law.
D) Energy Pattern Factor:
Energy pattern factor (EPF) is a useful parameter to
determine energy in the wind from the values of
mean wind speeds.
EPF=
∑
(ῡ)
Where ui=discrete wind velocity values (m/s)
ῡ=Mean annual or monthly wind velocity (m/s)
n=sample size
EPF values will be generally in the range of 1.5 to
2.50
E)Density of Air:
Density of airwill varywithchanges in altitude,
temperature and pressure. It is susceptible todiurnal and
seasonal variations also.The atmospheric pressure
decreases with height,the power output of wind turbines
on high mountains gets reduced compared to power that
could be produced at the same wind velocity at sea
level.
ρ=P/RT
-where ρ=Air density (Kg/m3
)
P=measured atmospheric pressure (mb)
T=Temperature of air in Kelvin ( ºc+273)
R=Universal gas constant
E) Turbulent intensity:
The short time perturbations of wind speed at a given
point in space over a time averaged mean value is
characterized by rapid changes in three dimensional
spaces.
T.I=(σ/ῡ)
Where σ=Standard deviation
ῡ=Mean wind speed
F) Weibull parameters:
Weibull distribution function is obtainedby performing
mathematical approximationson theactualvalues of wind
speed frequencies. It is defined by two-parameters
known as shape factor (ƙ) and scale factor (Ś).
f(ū)=(ƙS/Ś)(
ū
Ś
)ƙ
exp[-(ū/Ś)ƙ
] (ƙ>0,ū>0,Ś>1)
Where f(ū)=Probability density function
Ś=Scale Factor
ƙ=Shape Factor
= Standard deviation / mean wind velocity
Ś =1.12ῡ. Where, 1.5≥ƙ ≤3.0; ῡ=Mean wind speed
III. MODELLING OF WIND POWER
DENSITY USING ANN
Wind resource assessment & forecasting is a non-linear
mapping problem involving various parameters related to
climatic data and atmospheric modeling. For analyzing wind
resource assessment & forecasting problem, various
mathematical, probabilistic and statistical models
weredeveloped by researchers. These give predictions based
on extrapolation of data. But, an artificial neural network
(ANN) is a computing technique which has inherent ability
to interpret nonlinearities of prediction problem. ANN can
derive the required information directly from the
datasetswith fast response and generalization capabilities.
During training, ANNs will map the non-linearties of input
and output variables. After training the neural network for
certain amount of duration, complexities in input-output
mapping are understood by the network indicating the
stabilization of minimized error.Now, ANN can make
predictions or approximations of desired variables for new
values of inputs. Successful predictions will result if
mapping is perfectand complete. Radial Basis Networks is a
special technique of artificial neural network technology
which was adopted in function-approximation problems
related to wind energy resource assessment and wind power
forecastmodeling [1, 5]. In this work, radial basis neural
network is designed &developed for prediction &modeling
of wind resource data of target topographical locations.

_______________________________________________________________________________________
A. Data preparation & analysis
Variability in wind power resource greatly influences the
overall power plant output for megawatt range wind
projects.These fluctuations must be predicted accurately as
wind power distribution is highly nonlinear in
nature.National Institute of Wind Energy, Chennai formerly
known as CWET has established wind monitoring stations
as part of their wind energy survey programmes since 1985.
This work utilizes the wind resource mapping data bank of
76 wind monitoring stations under the published series viz.,
Wind Energy Survey – VIII. The collected data consisted of
measured wind parameters recorded during the years 1995
to 2011.In this work, wind data of 24 topographical
locations of Telangana and Andhra Pradesh states has been
utilized for modeling.In neural network modeling, problem
definition is a first step followed by systematic arrangement
of data. This step involves arranging all the wind parameters
data in time series chronology.As per the requirement of
ANN methods,data is processed using normalization before
being used in the modeling. This exercise involves filling
the missing values, smoothening of data and rescaling the
input output vectors. Most of the activation functionsof
neural networks demandthat the input-output values are in
specific range. Hence, input and output vectorsare
rescaledby dividing each dataset value with a scalar
constant. This process is known is data normalization. After
normalization, entire set of data patterns are divided into
training and testing data sets in the ratio of 80% and 20%. In
this problem, data pertaining to 19 stations is used for
training the neural network and data pertaining to 5 target
stations is used for testing or prediction phase.
B. Radial basis neural networks
Neural network modifiesthe values of certain variables with
respect to an input and output mapping. It begins with
initialization of weights associated with input& output
vectors and proceeds in the direction of finding a best fit of
the relationship.Thismodification of weights continues
during the feed forward movement of the neural network,
termed as epoch. Root mean square error (RMSE)is
evaluated in each training epoch and the network is run until
this error is minimized and stabilized. Many researchers
have designed and developed Multi-Layer perceptron
models for solving complex problems of large data
patterns.Back Propagation based Neural Networks have
already been used for predicting wind speeds for a new
topographical location [1]. Multilayer perceptron neural
network will have a nonlinear sigmoidal activation function.
Faster response of the neural network and better
approximation of the desired outputs is achieved through
radial basis function neural networks. Radial basis neural
network employs a linear learning rule that prevents the
networks to get struck up in local minima. It is an enhanced
algorithm giving improved accuracies by probabilistically
relating the outputs of each training epoch.
Radial basis neural networks are useful in performing the
exact interpolation of dataset points in any high-dimensional
space [17]. Thesewill have three layers- an input layer, a
hidden layer and a linear output layer. Hidden layer consists
of radial basis function implementation. At each value of
input of middle layer neuron, the radial distance between the
neuron centre and the input vector is determined. Radial
basis function is applied on this distance to evaluate the
output of the neuron in each epoch. Many radial basis
functions given belowhave been tried as activation
functions. A function S in linear space is defined as an
interpolation of radial basis vectors such that SR(xi) = ti, ; i =
1, …n.
And the basis function of the form is given by
ʋi (x) = ƭ (|| (x- xi)||)
Where ƭ is to map R+  R and the norm is Euclidean
distance. Radial basis functions tried here are listed below.
(i) Thin Plate Spline Function
ƭ ( ŕ ) = ŕ2
log (ŕ)
(ii) Gaussian approximation function
ƭ (ŕ ) = e – (
ŕ2
/τ2)
(iii) Multiple quadratic function
ƭ (ŕ ) = (ŕ2
+τ2
) 1/2
(iv) Inverse Multiple quadratic function
ƭ (ŕ ) = 1/ (ŕ2
+τ2
) 1/2
Where ŕ is the distance between from centre, τ is the width
of radial basis function.
C. SelectedArchitecture
To model the predictionproblem of wind power densitiesat
new locations, radial basis neural network is developed.
This consists of a hidden layer with built-in radial basis
function. Many trials were conducted by changing the
number of neurons in the hidden layers to find out the
optimum. 14-142-1 is derived to be the bestoptimal neural
networkarchitectureafterrunning the neural nets for various
combinations.
D. ANN Parameters
Fixing correct values for learning parameters, neural
network architecture is a herculean task and requires many
experimental trials. Speed of convergence, computational
accuracy, and neural network performance parameters are
importantin designing a best optimal network. Root Mean
Square Error is represented by ‘℮’and is calculated per
training iteration as given below:
------------- (1)
n
SR(x) = ∑ Wiʋi
i=1

_______________________________________________________________________________________
Table 1. Input- Output Mapping of wind resource
assessment & forecasting in neural network design
S.no. Input Parameter Output Parameter
1 Latitude
Mean Monthly Wind
Power Density ( WPD)
in Watt/ m2
2 Longitude
3 Altitude
4 Month
5 Height
6 Power Law Index
7 Energy Pattern
Factor
8 Peak Wind Speed
9 Year
10 Date
11 Prevailing Wind
Direction
12 Lull Hours of
percentage
frequency
Distribution
13 Air Density
14 Monthly Mean
Average Wind
Speed
Table 2.ANN architecture
S.no Algorithm ANN
Architecture
Activation
Function
1 Back
Propagation
Feed Forward
Neural Network
14-10-8-6-1
Input- Linear
Hidden-
Sigmoidal
Output-
Linear
2 Radial Basis
Function
Neural Network
14-142-1
Input- Linear
Hidden-
Radial basis
Output-
Linear
℮= ( ∑k | (pk – qk)| 2
) ½
------- (2)
Where pk is the actual target vectorof supervised learning
and qk is the output vector of each training epoch.
14-142-1 is finally selected as an optimum neural net
architecture after many trial and error exercises to run for
10000 training cycles. Learning rate isa constant whose
value is 0.6. Momentum isanother parameter for which 0.9
isfound to be suitable. Lesser value of learning rate slows
down running of neural network. But, a higher value for
learning rate enables the weights and error function to
diverge, thereby to stop learning at some point. A learning
rate of 0.6 is found to be optimal during this modeling
exercise. Table 2 lists the network architecture finally
designedas part of the modeling. Convergence of the neural
network program is linked to the fast response of the chosen
algorithm. Radial basis neural network is able to sense and
quickly understand the complexity of the data efficiently
than a traditional back propagation neural network.
E. Evaluating ANN results
Neural network tool box of MATLAB facilitates to validate
the data patterns during the evaluation phase. This is
performed to prevent over training. After this stage, testing
is performed by making predictions on new input datasets.
The predicted new wind power density values are compared
to the actual measured values of wind monitoring
stations.This is done as part of the regression analysis. Table
3 consists of predictions performedat target wind stations
endorsed by performance criterion of the neural nets. Other
important neural network performance parameters are
Coefficient of Determination (Dr) and Mean Average
Percentage error (M).These are calculated for both the back
propagationand radial basis neural nets using ORIGIN 6.0
software.
Dr= 1- (pk − qk)
M= (q-p)/p*100
Table 3. Neural Network training results
Village Name
Back Propagation Neural Net Radial Basis Neural Net
M R R2
M R R2
Kotrathanda 12.7068 0.9949 0.9898 40.3534 0.9987 0.9974
Kotturu 13.9735 0.9856 0.9714 70.6705 0.9953 0.9907
Shahapuram 11.8347 0.9986 0.9973 40.7337 0.9997 0.9995
Singarikonda 10.7298 0.9992 0.9985 60.7822 0.9990 0.9980
Tarennapallle 9.5281 0.9985 0.9970 80.6962 0.9990 0.9981
Total AP 11.7544 0.9954 0.9908 60.4472 0.9983 0.9967
Table 4. Predicted values of wind power densities
Back Propagation Radial Basis
Actual WPD in
Kg/m3WPD in Kg/m3
WPD in Kg/m3
56.61 57.49 58.83
89.86 94.90 94.75
101.46 100.51 103.16
79.70 81.89 84
79.04 78.18 82.5
81.33 82.59 84.65
Actual and predicted values for wind power densities under
changing topographical conditions are represented in Fig. 1.
This shows that neural network was able to predict the wind
power density value at the target location with at most
accuracy.Generalization capability of neural network
depends on the mode of training, the selection of training
data patterns covering most of the input-output variations
and activation function used. Here, Radial basis neural
network algorithm was successful in interpreting the non-
linear mapping between wind speed & other parameters and
wind power density at the target location.

_______________________________________________________________________________________
Fig: 1. WPD values predicted Vs actual
IV. WIND ENERGY GENERATION
ESTIMATES
Wind turbine power generation will not always follow
Manufacturer’s power curve. This exercise helps to
understand the overall energy generation that may be
possible at the selected area per year. Wind energy
generated at the new locations is estimated by analyzing the
manufacturer’s curves of horizontal axis wind turbines at the
average wind speed. Wind power density values are taken
from wind energy resource survey data.Power curves of
NORDEX 29 of 250 KW, ENERCON 33of 330kW,
Enercon82of 700 KW horizontal axis turbines are referred
for this purpose.Wind power density values estimated from
the modeling of radial basis neural networks along with
generation hours are mapped to the power curve equation to
estimate the wind energy generated in thatyear. This power
curve modeling exercise takes into account the performance
coefficients of the respective models of wind turbines given
by the manufacturer. Predicted wind power density values
along with the wind generation hours in that location
multiplied by swept area of the turbine give rise to the
energy units generated by the turbine. The graphs indicate
that in June and July, the number of units generated is
highest and contribute to 30% of overall energy that can be
generated in a year. The entire generation year may be split
up in to three regimes – low, high and moderate wind. Low
wind regime is during January to April. High wind season is
from May to August and moderate wind season is from
September to October. This variation pattern is almost
similar in Indian subcontinent.
Fig.2 Energy Predictions with Nordex N29 turbine
Fig.3 Energy Predictions with Enercon E48 turbine
Fig.4 Energy Predictions with Enercon E33 turbine
V. CONCLUSIONS
Wind power resource assessment and forecasting helps wind
farm developers to make rational decisions for the selection
of wind sites. Processing of topographical information along
with wind monitoring data with seasonal variations
underconsideration, will be crucial to developers during the
assessment of energy yield per season, month or year. These
variations will significantly influence wind energy
economics, energy trading,gridintegrationand transmission
requirements.Wind power forecasting is an important tool
for turbine manufacturers to select the wind power systems
with high efficiency and reliability. The developed RBFNN

_______________________________________________________________________________________
can be used to predict the wind power density at a target
topographical location.Itgives a better estimate of wind
power resource characterized by wind speed and other
important parameters at that specific location.In this work,
the deviations of wind power generation with
manufacturer’s power curve were also interpreted. The
performance of RBFNN is found to be satisfactory in
predicting the wind power densities across fivetarget
villages (Kotrathanda, Kotturu of the state of Telangana,
Shahapuram, Singarikonda and Teranapallle of the state
ofAndhra Pradesh) during the testing phase. Sensitivity
analysis was also performed to identify the effect of
influencing parameters viz., wind speed, direction, air
density, energy pattern factor, seasonal variation, turbulent
intensity etc.Predicted values of wind power densities are
utilized to map with power curves of wind turbines, to
estimate the number of units generated at eachtarget site if
that particular model of wind turbine is installed.
Thisresearch will go a long way inwind farm development,
turbine manufacturing and helps independent system
operators in finding risks of uncertainties due to variability
of wind source at selected locations.This work willalso be a
stimulus to wind energy market management in fixing a
proper value for energy priceindex.
REFERENCES
[1] Abdulkadir Yasar , Besir Sahin ,Mehmet Bilgili,
Application of artificial neural networks for the wind
speed prediction of target station using reference
stations data,Renewable Energy 32 (2007) 2350–2360.
[2] Aggarwa S.K., Meenu Gupta, Wind Power Forecasting:
A Review of Statistical Models, International Journal of
Energy Science (IJES) Volume 3 Issue 1, February
2013,www.ijesci.org
[3] Amol Phadke, Bharvirkar, Jagmeet , Khangura ,Ranjit,
International Energy Studies, Report on “Reassessing
Wind Potential Estimates for India: Economic and
Policy Implications”, Environmental
EnergyTechnologies Division,September2011;
http://ies.lbl.gov/India_Wind_Potential
[4] Anumakonda Jagadeesh, Institutional dynamics and
barriers in wind energy development A case study of
Tamil Nadu, and Andhra Pradesh, India, Center for
International Climate and environmental Research –
Oslo
[5] Badari Narayana.P, Dr.K.Hemachandra Reddy, Dr.S.
Srinivasa Rao, Development of distributed
topographical forecasting model for wind resource
assessment using artificial neural networks, Journal of
Sustainable Manufacturing & Renewable Energy,
ISSN: 2153-6821,Volume 2, Number 1-2© 2013 Nova
Science Publishers.
[6] Balakrishna Moorthy.C ,M.K. Deshmukh, ,Application
of genetic algorithm to neural network model for
estimation of wind power potential,Journal of
Engineering, Science and Management EducationVol.
2, 2010/42-48
[7] Bluff.J, McCullagh, K., and Ebert, E. - “A Neural
Network for Rainfall Estimation”,Conference Paper,
Proceedings of the 1995 Second New Zealand
International Two-StreamConference on Artificial
Neural Networks and Expert Systems, IEEE Computing
SocietyPress, Los Alamitos, Ca. U.S.A., pp. 389-392
[8] Bo Li ,Jie Dul andJin Wang,Li Ma, , Zhen Bin Yang, A
New Combination Prediction Model for Short-Term
Wind Farm Output Power Based on Meteorological
Data Collected by WSN, International Journal of
Control and Automation Vol.7, No.1 (2014)
[9] Boopathi K,Dr.S.Gomathinayagam ED/CWET, Lecture
Notes titled "India’s national initiatives and experiences
related to wind resource assessment"
Dr.S.Gomathinayagam Executive Director, K.Boopathi,
Scientist & Unit Chief i/c, WindResource Assessment
Unit Centrefor Wind Energy Technology, Chennai.
[10]Busawon K,L Dodson and M Jovanovic, Estimation of
the power coe¢cient in a wind conversion system,
Proceedings of the 44th IEEE Conference on Decision
and Control, and the European Control Conference
2005, Seville, Spain, December 12-15, 2005.
[11]Guang-Bin Huang ; Sch. of Electr. & Electron. Eng.,
Nanyang Technol. Univ., Singapore, Singapore ; Chee-
Kheong Siew, Real-time learning capability of neural
networks, Neural Networks, IEEE Transactions on
(Volume:17 , Issue: 4 ), July 2006, 863 - 878, ISSN:
1045-9227, DOI: 10.1109/TNN.2006.875974, IEEE
Computational Intelligence Society
[12]Indian Wind Energy Association; http://inwea.org
[13]Jeremy Parkes, Lecture notes on “ Taking the
guesswork out of Wind Power Forecasting, Jonathan
Collins- Husum 2012, , http://gl-garradhassan.com
[14]J.R. Parkes, Forecasting Short Term Wind Farm
Production in Complex Terrain, EWEC 2004.
[15]Ma, L., Luan, S.Y., Jiang, C.W., Liu, H L. and Zhang,
Y. (2009) A Review on the Forecasting of Wind Speed
and Generated Power. Renewable and Sustainable
Energy Reviews,13,915-920.
http://dx.doi.org/10.1016/j.rser.2008.02.002
[16]McSharry P. E., W. Taylor, Senior Member, IEEE, and
R. Buizza, Wind Power Density Forecasting Using
Ensemble Predictions and Time Series Models, IEEE
Transactions on Energy Conversion, 24, 775-782, 2009
[17]Powell, M. (1987), Radial basis functions for
multivariable interpolation: a review., in J. Mason & M.
Coxeds, ‘Algorithms for Approximation’, Clarendon
Press, Oxford.
[18]S.N Sivanandam, S. Sumathi, S.N. Deepa, Text book
Computer engineering series, title - Introduction to
neural networks using MATLAB 6.0
[19]T. Krishnaiah, K.Madhumurthy,S. Srinivasa Rao, K.S.
Reddy, Neural Network Approach for Modelling Global
Solar Radiation, Journal of Applied Sciences Research,
3(10): 1105-1111, 2007 © 2007, INSInet Publication.
[20]Wind Energy Resources Survey in India Vol. VIII
published by National Institute of Wind Energy, India
[21]Wind resource assessment in india, Dr. M. P Ramesh's
published lecture notes of CWET, India
[22]Wind resource assessment in india, E.Sreevalsan's
published lecture notes of CWET, India

Wind energy forecasting using radial basis function neural networks

More Related Content

What's hot

Similar to Wind energy forecasting using radial basis function neural networks

More from eSAT Journals

Recently uploaded

Wind energy forecasting using radial basis function neural networks