Symbolic regression using object-oriented genetic programming (1996)

Funding body: ERASMUS scheme (European Community)

Data driven modelling techniques have gained in popularity in the last 20 years. They are more cost effective compared to the development of mechanistic models. Furthermore, those mechanistic models are highly non-linear and complex, which makes them difficult to identify and use. Currently, the majority of data driven modelling methods can be categorised under two headings: artificial neural networks and statistical and regression analysis. Neural networks can usually provide models that are capable of good predictions, but they don't give any insight into the structure of the process. They are commonly called black boxes, one puts the data in and gets results from the model, but does not know anything about the underlying relationships between input and output data. It is usually desirable to gain some insight into the underlying process structures, as well as make accurate numeric predictions. The aim of this work is to develop a computer software, that uses genetic operations in order to find a symbolic equation describing the relationship between input and output data.

The structure and hence the complexity of the model or the equation is not specified like in the conventional regression, which seeks to find the best set of parameters for a pre-specified model. This new technique is called symbolic regression. Standard, bit-coded genetic algorithms are inadequate to represent varying sizes and shapes, hence the genetic programming technique is used.

References

  • P?yh?nen, H.O. and D.A. Savic, (1996), Symbolic Regression Using Object-Oriented Genetic Programming (in C++), Centre For Systems And Control Engineering, Report No. 96/04, School of Engineering, University of Exeter, Exeter, United Kingdom, p.72.

Back to Artificial Intelligence research and applications

Google+