2011 International Conference on Alternative Energy in Developing Countries and Emerging Economies
- 343 -
Abstract
-- This paper presents the short-term forecast of
wind speed along the coastal area at Chana district in
Songkhla province, southern Thailand. The observed wind
speed at 30 m height at Chana district in Songkhla province
during May until July 2010 was used to create the statistical
models in order to forecast the wind speed in August 2010.
Box-Jenkins and decomposition methods were applied for
comparing the model quality. Mean squared error was used
as main criteria for model selection. This criterion showed
that the Box-Jenkins method is better than the
decomposition method.
Index Term
s: Bayesian Information Criterion (BIC),
Box-Jenkins Method, Decomposition Method, Mean
Squared Error (MSE), Wind Speed.
I.
I
NTRODUCTION
Wind is one of the most difficult meteorological
elements to forecast. Observed wind speed and direction
is a result of the complex interactions between large
scales forcing mechanisms e.g. pressure gradient, Coriolis
Effect, and local characteristics of the earth surface [1].
There have been several works describing the probability
density function (p.d.f.) of wind speed thus the results
have shown that the p.d.f. of wind speed follows a
Weibull distribution in general. However, the p.d.f. of a
time series wind data does not convey any information
regarding to the sequential characteristics of the time
series which are crucial in forecasting [2]. The short-term
prediction involves the time scales in the order of some
days and from minutes to hours. Several researchers
proposed different approaches to forecast wind speed
time series. Literature review reveals that there are two
main schemes to forecast wind speed: the first is physical
model which uses many physical considerations to obtain
the best prediction precision. During the last two to three
decades, the numerical weather prediction (NWP) based
forecast and mesoscale models have been applied and
effective to predict the wind speed in large scale. NWP
model is developed by the meteorologist for large scale
area weather prediction using hydrodynamic atmospheric
models which consider physical phenomena. The second
is time series approach including traditional statistical
model. Physical method has advantages in long-term
prediction while statistical method performed well in
short-term prediction. Physical models use physical
consideration like terrain, obstacle, pressure, and
temperature to forecast the future wind speed.
According to the statistical methods introduced by
Box-Jenkins, these models can be divided as follows:
autoregressive model (AR), moving average model
(MA), autoregressive moving average model (ARMA),
and autoregressive integrated moving average model
(ARIMA) [3]. Recently, the state of the art based on
artificial intelligence like multi-layer perceptron (MLP),
Artificial Neural Networks (ANN), fuzzy logic, support
vector machine, some hybrid methods, as well as
wavelet-based methods were proposed and used
increasingly [4,5]. Based on statistical model, a pure AR
model was considered to model a series of data observed
in one month with daytime non-seasonality, and the non-
Gaussian shape of the wind speed distribution. Then,
ARMA model was used for a single series of one year
wind speed to forecast in short term period [6]. ARIMA
model was used in the formulation of the six years wind
data. Results showed that seasonal ARIMA models
present a better sensitivity to the adjustment and
prediction of the wind speed [4,7]. Wind power short-
term prediction was extensively reviewed for the recent
history [8]. This review gave a clear idea and lesson
learnt on the chronology and evolution of the short-term
prediction. Some useful topics were raised such as
adoption of a standard for measurement of performance
of models, improvement of the accuracy of existing
models and tools, methods which are able to provide
reliable estimates of the uncertainty of the predictions
form deterministic models, development of probabilistic
models, integration between mathematical/statistical and
physical/meteorological models for increasing the spatial
and time resolution, development of more accurate
upscaling and downscaling methods and new approaches
on complex terrain.
Among renewable energy sources, wind energy is one
of the greatest growths over the last decade. At the end of
2009, worldwide installed capacity of wind power was
159.2 GW with an energy production of 340 TWh,
accounted for 2% of worldwide electricity usage
compared to 0.1% in 1997 [5]. During the last three
years, the total amount of wind power has been doubled,
and this trend is expected to continue [9]. In Thailand, the
development of wind power increase considerably from
the order of kW in 1970s to MW in 2000s. In order to
utilize wind energy, the most significant problem to be
encountered is that wind has its own characteristics, i.e.,
Short-Term Forecast of Wind Speed
at Chana District, Songkhla Province, Thailand
W. Keerativibool*
,
**
,
***, J. Waewsak***
,
****, and P. Kanjnasamranwong*
,
***
*Department of Mathematics and Statistics, Faculty of Science, Thaksin University,
(Thailand)
**Centre of Excellence in Mathematics, CHE, Bangkok,
(Thailand)
, E-mail:
***Solar and Wind Energy Research Unit, Renewable Energy System Research and Demonstration Center,
(Thailand)
****Department of Physics, Faculty of Science, Thaksin University, Phatthalung,
(Thailand)
