You are on page 1of 45

System Identification using Particle Swarm Optimization based Artificial Neural Networks

Carlos J. Gmez-Mndez

University of f Puerto Rico at Mayaguez

Objective
Identify a system for which its mathematical model is

unknown Use Artificial Neural Networks (ANN) to model the system Train T the h ANN using Particle P l Swarm S Optimization(PSO) O (PSO)

System Identification
What is system y identification?
Mathematical model

Example:

System Identification
Why identify a system?
Complicated mathematical model Time consuming to develop the mathematical model

System Identification
Types yp of mathematical models:
White Box Model

Black Box Model

Artificial Neural Networks (ANN)


What is an ANN?
Mathematical model based on biological neural networks

Why ANN?
Adaptability Ad bili Non-linear characteristics

Artificial Neural Networks (ANN)


Model of a neuron:

Artificial Neural Networks (ANN)


Structure:

Particle Swarm Optimization (PSO)


What is PSO?
Swarm Intelligence Examples: Flocks of birds, School of fish Optimization algorithm

Why PSO?
Iterative

How it works?
Particles represent p set of data Example: X,Y space

Examples: Flock of birds

Examples: School of fish

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

Particle Swarm Optimization (PSO)

PSO Position Example


Target is (0,0): x 10 Particles
Random initial positions and velocities

Global l b lb best error and d local l lb best errors taken k into

consideration

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

2.5

1.5

0.5

-0.5 05

-1

-1.5

-2 2

-2.5 -2.5

-2

-1.5

-1

-0.5

0.5

1.5

2.5

Putting it all together


Necessary System data is measured ANN and PSO parameters are initialized Input data is fed to the ANN Error is calculated ANN is retrained using PSO Repeat until acceptable error is achieved

Overview

System to be identified
The transfer function of the system being modeled with an ANN

is:

The input p to the system y is an unit step: p

Results
System response to unit step input and output of initialized

ANN:

Results
System response to unit step input and output of trained

ANN:

Future Work
The algorithm developed will be applied for maximum

power tracking of an ocean wave energy conversion system. Preliminary simulations will be done in real-time using dSPACE/SIMULINK and the setup shown. shown
The algorithms will be used f power management of for fa DC Motor.The setup consists of a DC Motor, DC Generator, Power Converter Board, DSP Board, and PC.

Questions?

You might also like