forecast of long term wind speed based on optimized support vector regression (1)
TRANSCRIPT
![Page 1: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/1.jpg)
Long-Term Wind Speed Prediction based on
Optimized Support Vector Regression
by
Sarah Osama Teaching Assistant, Department of Computer Science
Faculty of Computer an Information
Minia University
http://www.egyptscience.net
![Page 2: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/2.jpg)
Agenda 2
Introduction
Wind Speed Problem
Related Work
Problem Definition
Proposed WOA-SVR Algorithm
Experimental Results and Discussion
Conclusion and Future Works
![Page 3: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/3.jpg)
Introduction
3
The 8th IEEE International Conference on Intelligent Computing and Information Systems (ICICIS 2017)
Wind energy one of the fastest rising energy resources are wind energy since it is
renewable, plentiful and clean. One of the most important main sources of
renewable energy is the energy of the wind. One of the drawbacks for the reliability
and precision of the power system is the fluctuation and nonlinear of wind
![Page 4: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/4.jpg)
Introduction
4
????
Wind speed prediction is classified into three categories, named as,
Short term prediction could point to forecasting data for a few hours
ahead no more.
Medium term prediction is for a duration ranging from about a few
hours to three days
Long-term prediction point to duration more than three days ahead;
yet, there is no ultimate end of the duration
![Page 5: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/5.jpg)
5
Introduction
Algorithms Used for Wind Speed Prediction
Structural break model and Bayesian
theory Neural Network
Support Vector Regression
(SVR)
![Page 6: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/6.jpg)
6
Introduction
SVR
![Page 7: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/7.jpg)
7
Introduction
Linear data in 2D
(Discrete Data)
Linear data in 2D
(Continuous Data)
![Page 8: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/8.jpg)
8
Introduction
Non-Linear data in 2D
(Discrete Data)
Non-Linear data in 2D
(Continuous Data)
![Page 9: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/9.jpg)
9
Introduction
Non-Linear data in 2D
(Discrete Data)
Non-Linear data in 3D
(Discrete Data)
Kernel Function
(i.g RBF)
Input space Feature space
![Page 10: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/10.jpg)
10
Introduction
SVR
![Page 11: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/11.jpg)
11
Introduction
Overfitting Fit
(Good generalization) Underfitting
Very small γ or
Very large d
Very large γ or
Very small d
Optimal value
of γ or d
![Page 12: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/12.jpg)
12
Introduction
SVR
![Page 13: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/13.jpg)
13
Introduction
SVR
![Page 14: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/14.jpg)
Introduction
14
Optimization Algorithms
Conventional algorithms Meta-heuristic optimization
algorithms
Individual-based Optimization Algorithms
Population-based Optimization Algorithms
Evolutionary Algorithms Swarm Optimization
Algorithms
![Page 15: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/15.jpg)
Introduction
15
![Page 16: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/16.jpg)
Introduction
16
Pos= [5,3] Pos= [5,5]
Pos= [1,5]
Pos= [2,5] Pos= [1,6]
Pos= [8,7] Pos= [3,7]
Pos= [8,8]
Pos= [4,5]
Pos= [1,6]
Pos= [1,7]
Pos= [1,8]
Pos= [6,5]
![Page 17: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/17.jpg)
Introduction
17
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 18: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/18.jpg)
Introduction
18
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc =66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 19: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/19.jpg)
Introduction
19
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Pos= [1,6]
Acc = 63%
Acc =50%
Acc =53%
![Page 20: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/20.jpg)
Introduction
20
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 21: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/21.jpg)
Introduction
21
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 22: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/22.jpg)
Introduction
22
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 23: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/23.jpg)
Introduction
23
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 24: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/24.jpg)
Introduction
24
Acc = 62% Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 25: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/25.jpg)
Introduction
25
Acc = 62%
Acc = 60%
Acc = 64%
Acc = 55% Acc = 66%
Acc = 60% Acc = 59%
Acc = 60%
Acc = 60%
Acc = 80%
Acc = 63%
Acc =50%
Acc =53%
![Page 26: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/26.jpg)
Introduction 26
Using Whale Optimization Algorithm(WOA) to
Optimize Kernel function parameter
Optimize Penalty parameter
Feature selection
Comparing our results with Traditional SVR and PSO
![Page 27: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/27.jpg)
27
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
1) In the collected data set, there are
some problems such as missing
values. To overcome this problem,
linear interpretation is used to fill
these gaps and solve this problem.
Proposed WOA-SVR Algorithm
![Page 28: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/28.jpg)
28
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Proposed WOA-SVR Algorithm
![Page 29: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/29.jpg)
29
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
The dataset is divided into two
subsets namely training and testing
datasets using k-fold cross
validation algorithm(K=3)
Proposed WOA-SVR Algorithm
![Page 30: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/30.jpg)
30
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Proposed WOA-SVR Algorithm
![Page 31: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/31.jpg)
31
WOA use mean abstract
percent error (MAPE) to
determine the goodness or
quality of each search agent
or solution
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Proposed WOA-SVR Algorithm
![Page 32: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/32.jpg)
32
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
WOA Generate initial
population randomly
according to the
available ranges
Proposed WOA-SVR Algorithm
![Page 33: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/33.jpg)
33
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Train SVR using training
dataset and selected
parameters
Proposed WOA-SVR Algorithm
![Page 34: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/34.jpg)
34
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Test SVR model using
testing dataset
Proposed WOA-SVR Algorithm
![Page 35: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/35.jpg)
35
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Calculate fitness function for
each whale and select the
best one to be the optimal
solution
Proposed WOA-SVR Algorithm
![Page 36: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/36.jpg)
36
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
When the termination criteria are
satisfied, the operation ends;
otherwise, we proceed with the
next generation operation. In the
proposed algorithms, the WOA
is terminated when a maximum
number of iterations are
reached.
Proposed WOA-SVR Algorithm
![Page 37: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/37.jpg)
37
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Return SVR optimal parameters
and prediction Model
Proposed WOA-SVR Algorithm
![Page 38: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/38.jpg)
38
Wind Dataset
Data Preprocessing Initialize WOA Parameters
Define the initial cost function Preprocessed
Dataset
Training
set
Testing
set
Train SVR classifier
SVR Model
Test SVR model
Satisfying stopping
criterion ?
Generate New Solution
Optimized Parameters (C, γ
and features subset) Prediction Model
Yes
SVR Parameters
(C, γ and features subset)
SV
R P
ara
mete
rs
(C, γ a
nd
featu
res sub
set)
NO
Generate initial population randomly
according to the available ranges
Each whale proceeds to its
next locations.
Proposed WOA-SVR Algorithm
![Page 39: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/39.jpg)
Experimental results and discussion 39
Description of Experimental Data
The used data in this paper is measured on an average
daily basis from a meteorological station at Helwan
University, Cairo, Egypt, these data were gathered for the
year 2016.
In this paper, sixteen meteorological factors
![Page 40: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/40.jpg)
Experimental results and discussion 40
Description of Experimental Data
The time of the study ranges from January 1, 2016, to
December 22, 2016, except some missing days. The data
set includes 197 wind samples
These data are complete and daily average weather data
that collected and employed from the Space Weather
Monitoring Center (SWMC) in Egypt
![Page 41: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/41.jpg)
Experimental results and discussion 41
Evaluation metrics
Convergence
Forecasting Accuracy Measurements
MAE
MSE
MAPE
RMSE
![Page 42: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/42.jpg)
Experimental results and discussion 42
Comparison of Convergence:
![Page 43: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/43.jpg)
43
Experimental results and discussion
Forecasting Accuracy Measurements
![Page 44: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/44.jpg)
Conclusion 44
In this paper, a novel optimization algorithm based on the Whale
Optimization Algorithm (WOA) and SVR called WOA-SVR was proposed to
optimize the SVR parameters. This algorithm is employed to enhance the
performance of SVR in order reduce the regression for wind speed
prediction.
The performance of WOA-SVR algorithms is estimated and compared using
wind datasets as real world cases.
From the experimental results, it can be concluded that WOA-SVR algorithm
achieved MAPE, MAE, MSE and RMSE lower than PSO-SVR and traditional
SVR.
![Page 45: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/45.jpg)
Future Works
Multi-kernel learning SVR for wind speed prediction,
Chaotic swarm optimization for SVR parameters optimization,
Present a comparative analysis of a swarm optimization
algorithms and Kernel functions
It is an important issue to check the reliability and stability of
proposed algorithm, and this can be done by increasing the
dataset size.
45
![Page 46: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/46.jpg)
46
Questions!?
![Page 47: Forecast of long term wind speed based on optimized support vector regression (1)](https://reader033.vdocuments.us/reader033/viewer/2022052606/5a6597ed7f8b9ab3488b4667/html5/thumbnails/47.jpg)
Thanks and Acknowledgement 47