Solar Power Prediction Based on Satellite Images

and Support Vector Machine

Summary:-

Whole idea of this project is solar power prediction model based on a large amount of
various historical satellite images and an SVM learning scheme. They analyzed 4
years historical satellite images and utilized them to congure a large number of
inputs and output data sets for the SVM learning. This project was done in South
Korea.
There have been several studies on prediction technologies for solar irradiance or PV
power generation. But they categorize them into three major categories.
1. solar irradiance or cloud index prediction technologies based on satellite
images
2. cloud detection technologies for solar irradiation prediction
3. solar power prediction technologies based on machine learning schemes
In this paper, they propose a solar power prediction model based on various satellite
images such as atmospheric motion vector (AMV), clouds, and irradiance images, and
an SVM learning scheme.
The proposed SVM-based prediction model can simultaneously predict the future
amount of clouds and solar irradiance on multiple sites of South Korea in the range of
15 to 300 minutes (intra-day).
They show the prediction accuracy of the proposed model, compared with that of the
conventional time-series and the ANN models.
Satellites images utilize in this paper are briefly explained below.

- Atmospheric motion vector (AMV): provides the overall information of atmospheric

motion and wind direction and speed of lower, middle, and upper wind elds

-Clouds analysis images: provides a variety of cloud analysis images, which contain
the amount of clouds, cloud shape, and cloud thickness images

- Irradiance images: based on the amount of light intensity reected by the ground
and the irradiance is measured in Watt per square meter (W/m2) in the range of 0
1000 W/m2

Under the PROPOSED AMV EXTRACTION METHOD, they explained the whole
method in four steps in order to gather information of wind direction and speed.
Steps: The rst step is to separate red, green, and blue pixels from an AMV
satellite image and save their pixel coordinates into each repository of red AMVs,
green AMVs, and blue AMVs.

The second step is to extract each single blue AMV from the repository of blue
AMVs. (Mention blue only because for the easiness)
 The third step is to nd the head, body, and tail pixels from the sub-repositories of
the single blue AMVs.

The fourth step is to calculate the wind speed for each single blue AMV.
The topic EXTRACTION OF CLOUD IMPACT FACTORS FROM SEARCH
AREAS is used for explain the calculation of cloud impact factor step by step.
Their solar power prediction model based on the support vector machine (SVM)
regression scheme can be trained with a large amount of historical satellite image
data.
- Support vector machine (SVM) is one of high-performance ML schemes
-Training method has two objectives of prediction. The rst one is to predict the
future amount of clouds at the target area and the other one is to predict the future
irradiance at the target area.
Then evaluated the performance of the proposed SVM-based prediction model.
According to that SVM-based prediction model shows the highest prediction
accuracy, compared to other prediction models such as the conventional time-series
and the ANN models.
In this article Jang et al. give better description of solar power prediction model based
on a large amount of various historical satellite images and an SVM learning scheme.
The path for the satellite images and an SVM learning scheme is explained step by
step in this article. The authors used metrological satellite images for North-East Asia
from Korea National Meteorological Satellite Center (NMSC) to present the
extraction method of clouds factors and to train solar power prediction model based
on the support vector machine (SVM) regression scheme. Their research focus on
analyze 4 years historical satellite images and utilize them to congure a large
number of input and output data sets for the SVM learning. The performance of the
proposed SVM-based prediction model is also evaluated in this research. This article
is useful to our research topic in order to understand the prediction methods for solar
power regarding satellite images and SVM and the performance of the proposed
method. The main limitations of this article are prediction was done using 4 years
images and performance is introduced by comparing with time-series model and
artificial neural network (ANN). Author try to explain the prediction based on satellite
images and an SVM learning scheme and also its performance for various time
horizons and radii of the target area with comparing time-series model and ANN. This
article will not the basis of my research; however it is useful to get an idea about
satellite image processing.