NEURAL NETWORK
By : Farideddin Behzad Supervisor : Dr. Saffar Avval
May 2006 Amirkabir University of Technology
Agenda
Definition Application fields History Application Biological inspiration Mathematical model Basic definition Learning Neuron types and some issues Example of application in energy & engineering
2
Definition
Haykin(1999) massive parallel-distributed processor natural propensity for storing experiential knowledge available for use. Acquiring knowledge by the network from its environment through a learning process Using interneuron connection strengths, (a.k.a. synaptic weights), to store the acquired knowledge
3
Application fields
Data analysis Pattern recognition Control application
4
History
1943, Warren McCulloch & Walter Pitts, works of neurons 1960, Bernard Widrow & Marcian Hoff, developed ADALINE and MADLINE From late 1960s to 1981, decreasing of researches Early 1980s, renewed interest in neural network 1986, Daivid Rummelhart & James McLand, error backpropagation algorithm
5
Applications
Aerospace industry Automotive industry Banking Military industry Economics Manufacturing Medical applications Oil & petroleum industry And many more …
6
Biological inspiration
Brain structure Cell
Cell
Dendrites
body Axon Denderites
Soma (cell body)
Axon
7
Mathematical model
x1 x2 x3
… xn-1 w3 w2 w1
z
Node
w x ; y H (z)
i i i 1
n
Output y
Inputs
. wn-1 wn
Artificial neural cell
8
xn
Mathematical model
Mathematic model of artificial neural cell
Cell body
input
f
wp b
p
b
9
output
w
n
a
Basic definition
Architecture: formal mathematical description of a Neural Network. (feed-forward & feed-back) Layer or Slab: A subset of neurons in a neural network. (Input, Hidden, Output) Episodical vs continuous networks Neuron weight Activation function
10
Activation function
Linear
Activation function
Non-Linear
Step
Sigmoid
Linear
Gaussian
11
Learning
Supervised learning
learning
Unsupervised learning
Coincidence learning Performance learning Competitive learning Filter learning Spatiotemporal learning
12
Neuron types
Hebb Perceptron Adaline Kohonen
13
Some issues
Training dataset Test dataset Network size
14
Example of application in energy
Soleimani. M, Thomas. B, Per Fahlen, “Estimation operative temperature of building using artificial neural network”, Journal of Energy and Building 38 ,2006 Luis M. Romeo, Raquel Gareta, “ neural network for evaluating boiler behaviour”, Applied Thermal Engineering 26, 2006 Seyedan B., Ching C.Y., “Sensitivity analysis of freestream turbulence parameter on stagnation region heat transfer using a neural network”, International Journal of Heat and Fluid Flow, 2006 Perez-roa P., Vesovic V., “Air-pollution modelling in an urban area: Correlation turbulent diffusion coefficients by means of an artifical neral network approach”, Atmospheric Environment 40, 2006
15
References
.1 2.
Hecht-Nielsen R., “Neurocomputing“, publishing company, 1991
Addison-Wesley
3.
MATLAB help documentation
16