You are on page 1of 5

Classical v/s Evolutionary

Optimization Methods

Rajkumar S. Pant
Associate Professor
Aerospace Engg. Deptt.

AE- 456/755 Optimization in Design


Definition of Classical Methods
Point-by-point methods
Single solution update in each iteration

Deterministic Transition Rule


Unique mathematical formula

Two basic types


Direct
No derivates used

Gradient based
First and/or second order derivates used
AE- 456/755 Optimization in Design
Std. Classical approach

Start from a random starting point

Locate a search direction


Transition rule
Mostly using local information

Carry out a unidirectional search

Check convergence criteria

Repeat
AE- 456/755 Optimization in Design
Direct v/s Gradient based
Direct Gradient based
Slow convergence Fast Convergence
Many function Lesser function
evaluations evaluations
Same code for different Cant be used for non-
problems differentiable or
discontinuous functions

Common problems
•Lack robustness
•Premature convergence
•Lack generality
•Difficult to handle discrete spaces
•Cannot be parallelized

AE- 456/755 Optimization in Design


Evolutionary Optimization
Work on a population of designs
Controlled randomness in search
Some examples
Genetic Algorithms
Simulated Annealing
Evolutionary Programming
Genetic Programming

AE- 456/755 Optimization in Design