Misc,
Genetic Programming Techniques Applied to Measurement Data
Diploma Thesis, (February 1997)
Abstract
In this project, an unknown system structure was
identified using the Genetic Programming technique. A
program was developed that, instead of combining linear
systems, evolves nonlinear ordinary differential
equations to describe a system. The solution space was
increased by using this approach. Automatically defined
functions were used to represent the ordinary
differential equations.
A new way was introduced to speed up the evolution of
genetic trees. The genetic trees representing the
equations were written in C notation to a file to be
compiled by a C compiler and evaluated by the computer.
This required modifications to the Genetic Programming
kernel formerly used. The modifications facilitated
evaluation of a complete generation at one time to
minimise compiler overhead. In order to build up a C++
class hierarchy, the kernel was completely
restructured. A lot of features like shrink mutation,
variable tournament size, improved deme handling etc.
were also added. A parameter study was carried out to
investigate the influence of important control
parameters.
The Genetic Programming system was applied to a known,
nonlinear system to verify its ability. After this
proved to be successful, the input/output response data
of a helicopter engine was used to identify that
system. Candidate models were derived from that
analysis.
Tags
- algorithms,
- c++
- class
- engine
- genetic
- helicopter
- identification,
- library,
- programming
- programming,
- system
