2. Outline
Introduction and Background
Objectives
Load Forecasting Methods
Why ANN?
Proposed Approach
What Are Neural Networks
Different Types of Neural Networks
Network Structure
Training a Neural Network
Back Propagation
Simulation Results
Training Performances
Result Comparison
Conclusions
3. Introduction and Background
Objective
Electric power generation, transmission, distribution, security
Increase or decrease output of generators
Interchange power with neighboring systems
Prevent overloading
Electric power market
Price settings
Economic operation of power plants
5. What Are Neural Networks?
Massively parallel networks of simple processing elements (neurons)
Designed to emulate the functions and structure of the brain
can solve very complex problems
a new method of programming computers
6. Different Types of Neural Networks
Feed-forward Network
Signals travel in one way only; from input to output
No feedback
the output of any layer does not affect that same layer
Feedback Network
signals traveling in both directions by introducing loops
Feedback networks are dynamic
They remain at the equilibrium point until the input changes
7. Network Structure
Estimate the number of layers and of neurons
trial and error procedure
Two types of adaptive algorithms can be used:
start from a large network
begin with a small network
8. Network Training
The process of determining the network parameters to achieve the
desired objective
Neural networks learn from examples
The most basic method of training; Trial and Error
epoch-by-epoch learning
The Aim: determine a set of weights which minimizes the error
9. Back Propagation
Most widely and frequently used neural network learning algorithm
mathematically designed to minimize the error
and propagate backward the local error terms
10. Simulation Results
Forecasting Procedure
Data Source
Ontario weather stations and dispatching centers
Historical Data
Load – load for the year 2008
Weather – weighted average hourly weather conditions of
stations in Ontario Province, Canada for 2 years
12. Result Comparison
The network test simulation should be made in order to find out the
performance in a real problem:
The simulation result of the trainbr algorithm with 8 neurons
13. Result Comparison
The simulation results of both trainbr and trainlm with 8 neurons
The simulation results of trainlm with 8 neurons and 30 neurons
Test target, Trainlm with 30 neurons, trainlm with 8 neurons
14. Conclusions
Trainbr is one of the best choices to do load forecast
More neurons are needed in network structure to obtain
accurate results
A few of the simulation part didn’t match the real demand
because of lack of information
More enough information and a precise training give us better
results for load forecasting in a smart grid.