Apidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
West denmark short term load forecast_for smart grids
1. Machine Learning
Short-term Load Forecasting in the Electrical Grid
Alexandru Ceocea
aceoce12@student.aau.dk
Mohammed Seifu Kemal
mkemal11@student.aau.dk
Robin Doumerc
rdoume12@student.aau.dk
NDS9
Department of Electronic Systems
Aalborg University
Denmark
2. Agenda
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Short Term Load Forecasting
Data Collection
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Learning Algorithms
Linear Regression
Neural Networks
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs Neural Networks
Conclusions
Conclusions
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
3. Smart Grid Networks
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
2
What is a Smart Grid ?
Smart Grid Networks
Short Term Load
Forecasting
Modernized electrical grid that makes use of information and
communication technology in order to gather and react on
information such as the behavior of suppliers and consumers
in an automated centralized way
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
4. Smart Grid Networks
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
2
What is a Smart Grid ?
Smart Grid Networks
Short Term Load
Forecasting
Modernized electrical grid that makes use of information and
communication technology in order to gather and react on
information such as the behavior of suppliers and consumers
in an automated centralized way
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Why Smart Grids ?
Linear Regression vs
Neural Networks
To improve the efficiency, reliability and sustainability of the
production and distribution of electricity within the Grid.
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
5. Short Term Load Forecasting
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Load Forecasting
Introduction
Smart Grid Networks
3
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
6. Short Term Load Forecasting
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Load Forecasting
Introduction
Smart Grid Networks
Vitally important for the electric industry
3
Short Term Load
Forecasting
Data Collection
Balance supply and demand
Learning Algorithms
Infrastructure development
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
7. Short Term Load Forecasting
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Load Forecasting
Introduction
Smart Grid Networks
Vitally important for the electric industry
3
Short Term Load
Forecasting
Data Collection
Balance supply and demand
Learning Algorithms
Infrastructure development
Linear Regression
Neural Networks
Results
Linear Regression
Short term Load Forecasting
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
8. Short Term Load Forecasting
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Load Forecasting
Introduction
Smart Grid Networks
Vitally important for the electric industry
3
Short Term Load
Forecasting
Data Collection
Balance supply and demand
Learning Algorithms
Infrastructure development
Linear Regression
Neural Networks
Results
Linear Regression
Short term Load Forecasting
Neural Networks
Linear Regression vs
Neural Networks
From 1 hour to 1 week
Conclusions
Generation of short term scheduling functions
Conclusions
Assessing the security of the power system
Dispatcher information
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
9. Collected Data
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Training data is composed of energy consumption measured
over the course of one year (2011), in West Denmark and is
provided by Energinet.
Short Term Load
Forecasting
4
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Date
Results
Energy consumption (MWh)
Linear Regression
Hourly update
Linear Regression vs
Neural Networks
Neural Networks
Conclusions
Time frame = 1 year
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
10. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
5
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
11. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
12. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
x1 - Day of the week
Linear Regression vs
Neural Networks
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
13. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
x1 - Day of the week
Linear Regression vs
Neural Networks
x2 - Day of the month
Conclusions
Conclusions
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
14. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
x1 - Day of the week
Linear Regression vs
Neural Networks
x2 - Day of the month
Conclusions
Conclusions
x3 - Average previous load (24h)
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
15. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
x1 - Day of the week
Linear Regression vs
Neural Networks
x2 - Day of the month
Conclusions
Conclusions
x3 - Average previous load (24h)
x4 - Load of same time frame (1h) on previous day
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
16. Linear Regression
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
For the forecasting of electric load consumption, regression is
used to model the relationship between the load and similar
characteristics from a previous time frame.
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Regression formula used: hθ (x) = θT x =
5
n
Linear Regression
Neural Networks
θi xi
Results
i=1
Linear Regression
Neural Networks
x1 - Day of the week
Linear Regression vs
Neural Networks
x2 - Day of the month
Conclusions
Conclusions
x3 - Average previous load (24h)
x4 - Load of same time frame (1h) on previous day
x5 - Load at same time, same day, previous week
x6 - Month
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
17. Neural Networks
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
6
Neural Networks
Results
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Figure: Artificial Neural network
Conclusions
Same features as before
Comparison purposes
Better data fitting
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
18. Linear Regression - 4 features vs 6 features
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
7
Linear Regression
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
Figure: 24 Hour prediction using Linear Regression
MAPE4ft = 8.060
MAPE6ft = 8.473
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
19. Results Neural Networks
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
8
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
Conclusions
Figure: 24 Hour prediction using Neural Networks - 6 features
MAPE = 5.060
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
20. LR vs NN - 6 features
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
9
Linear Regression vs
Neural Networks
Conclusions
Conclusions
Figure: Linear Regression vs Neural Networks
MAPELR = 8.473
MAPENN = 5.060
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
21. Conclusions
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Linear Regression
Smart Grid Networks
Short Term Load
Forecasting
Data Collection
Learning Algorithms
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Neural Networks
Linear Regression vs
Neural Networks
Conclusions
10
10
Conclusions
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark
22. Conclusions
STLF
A. Ceocea,
M.S. Kemal,
R. Doumerc
Introduction
Linear Regression
Smart Grid Networks
Short Term Load
Forecasting
More features = better training data fitting
Data Collection
Learning Algorithms
Validation data fitting might not be optimal because of the
non linearity of the system
Linear Regression
Neural Networks
Results
Linear Regression
Neural Networks
Neural Networks
Better adapted to non-linear systems
Linear Regression vs
Neural Networks
Conclusions
10
Conclusions
Better overall results based on our implementation
10
NDS9
Dept. of Electronic Systems
Aalborg University
Denmark