Improving engineering design models using an alternative genetic programming approach

摘要

This paper describes an alternative approach to Genetic Programming (GP) for engineering design model development. The algorithm is initially developed to solve Boolean induction and simple symbolic regression problems within a discrete searchspace. This technique, called "DRAM-GP" (i.e. Distributed, Rapid, Attenuated Memory GP), is based upon a steady state population utilising a novel constrained complexity crossover operator. Node complexity weightings are introduced to provide a basis forspeciation. Separate species of solutions, classified by complexity can be established which act as discrete GP sub-populations communicating with each other via crossover. The technique is extended to incorporate both continuous and discrete searchspaces (HDRAM-GP" i.e. Hybrid DRAM-GP). HDRAM-GP includes a real numbered Genetic Algorithm (GA) to aid search in the continuous space. Its application is demonstrated on engineering fluid dynamics systems.