A Versatile Genetic Algorithm System in Ada: Implementation and Applications
Max Edwards
The following points summarise the aspects that are lacking in VAGAS as it is presented in this report, and suggest additional development work that may improve its usefulness and versatility.
• Additional work is required on the implementation of the time tabling problem, in order to make it more responsive to optimisation techniques such as genetic algorithms.
• The software contains system dependencies, in the Basic_Maths, Random_Numbers and Graphics packages. Some time spent implementing portable versions of these packages would considerably improve the system's potential usefulness.
• Implementation of many more new and unusual crossover, mutation, selection and scaling operators is required.
• There is room for improvement in the VAGAS user-interface; the provision of a complete graphical interface would greatly increase the system's attractiveness and usability.
• Further experimentation with the DeJong test functions should be carried out, in order to judge the effectiveness of genetic algorithms against other optimisation methods.
The research work carried out in this project raised the following issues, which were not followed up due to lack of time:
• Further study is required of the effects and usefulness of periodic population re-seeding ; a facility for doing this has already been installed in the system (see 4.8).
• Further comparison between the various mutation operators is required.
• There is scope for much more experimentation with the Meta-level GA parameter-set optimiser, and it could be applied to other problems, or in such a way as to determine a parameter-set which performs well over all problems; this in itself holds potential for results of considerable importance.
• The GA performance rating that was proposed in 4.2. should be investigated.