You are on page 1of 12

Artificial Intelligence Based Technique for Electrical

Energy Demand Prediction

Dr. Badar ul Islam


Head Department of Electrical Engineering
NFC-IEFR Faisalabad
Introduction: Electrical Load Forecast Process

Forecast Results

Forecast Model

2
Introduction: Categories of load forecast

Area of interest

Short term load forecast (STLF)


- With a lead time of up to a few hours or days ahead
- Emphasized for the economic and day to day operational
decisions of systems
Medium term forecast
- With a lead time of few weeks to one year period
- Deals with the scheduling of fuel supplies and maintenance
operations

Long term forecast


- With a lead time of up to ten years
- Useful for planning operations of the power systems

Lead Time Horizon


3
Introduction: Why Short Term Load Forecast (STLF) is
Important?

• Electric Power Sector is the major source of Carbon Dioxide emission, that
pollutes the environment.
U.S. Energy Information Administration /
Monthly Energy Review November 2012 Accurate STLF can significantly
DOE/EIA-0035(2012/11)
reduce
CO2 Emission From Energy Consumption
www.eia.gov/mer the unnecessary
(Million Metric Tons)emission
Jan-Decof2020
Carbon Dioxide, that leads to save
the environment.

1485 U.S. Energy Information


19.65 % 1850 Administration / Monthly
24.47 % Residential Sector Energy Review November
994 Commercial Sector
13.15 % 2012
Industrial Sector
Transportation Sector DOE/EIA-0035(2012/11)
Electric Power Sector
1160
15.36 % www.eia.gov/mer
2170
28.71 %

4
Introduction: Why STLF is Important?

• Accurate STLF can cause a vital impact on the economy of a country by saving
the money and other resources. Can cut down the
fuel cost

It has been
estimated that
Can cut down the an increase of
labour cost 1% in the
forecast error
caused an
increase of 10
million pounds
in operating
Can reduce the cost per year for
an electric
maintenance cost utility in the
Model B results United
Model A results
Mean Absolute Kingdom [13].
Mean Absolute
Percentage Error Percentage Error
of 5%. of 4%.
5
Introduction: Why STLF is Important?

• It is indispensable to the reliability of power system operations.

Prevents
load
shedding

Useful in
scheduled Prevents
maintenance equipment
of plant
failure and
blackouts

6
Introduction: Existing Techniques for STLF

Energy Demand
Forecasting Techniques

Parametric Non Parametric


Techniques Techniques

Time Series Exponential Regression Based Artificial Neural Fuzzy Ligic Expert Systems
Techniques Techniques Techinues Network

Non Parametric
Parametric Techniques Hybrid Techniques
Techniques

PROS:
PROS
• Input and output mapping ability
• Transparent Modelling techniques • provide the combined advantages of
• Massively parallel networks
• Physical Interpretation between the used techniques
• Lesser programming and
inputs and outputs. • Generally apply probabilistic rules not
mathematical formulation
deterministic ones.
• Robust behavior to deal with non
CONS
• Limited ability to address the non linear parameters
CONS:
linear load behavior.
• Dependence on initial parameters
• Complex mathematical models.
• Long training time
• Heavy computational effort
• Improper choice of network topology
• Inefficient convergence

7
Gradient Descent Algorithm (GDA)
Calculation of error gradient and weight adjustment at one point in ANN

• The error function of the GDA used in BP (6)


network for each training pattern can be Since, derivative of the error with respect to
expressed as [82]; the output error of a neuron j can be
expressed as; ,
(1) (7)
(2) Hence, the error gradient w.r.t. one
where, Ep is the error at point p, to an yo are particulate weight can be expressed as;
the predicted and actual outputs at point p. (8)
½ is a mathematical balance justification of The change in weight is described as;
the square of the difference of the error. (9)
• The error gradient at point p is • The value of the new weight after
(3) adjustment can be described as;
(4) (10)
• By applying the chain rule of • where, is the learning rate, is the
differentiation on Equation (4) change in weight, is the new weight and
(5) represents the error.
• From the Equation (2), following
expression can be extracted;
Back-Propagation (BP) Algorithm
Calculation of error gradient and weight adjustment at one point in ANN

Forward Pass of BP Back Pass of Error Gradients of BP


Backpropagation

12

You might also like