An Evaluation of Significant Genetic Operators (as applied to the Travelling Salesman Problem)

my fourth year project involved studying genetic algorithms. i used the tsp to study the way operators affected the algorithm - rates of convergence, success in reaching an optimal solution, genetic drift and so forth. my supervisor was George Mitchell, who is working with GA's for his Phd.

for this project, i wrote a genetic algorithm (called evolver) to solve the tsp. the code for this is available from the downloads page. it's written in java, and requires jdk 1.3.1 or better to compile and run. the code is packaged as a gzip'ed tarball with javadoc-generated documentation and a readme.txt file.

some results are contained in the plots file on the same page as the code. these are comma-delimited files of the best individual from every 10,000th generation. each run lasted for 500,000 generations, with 150 individuals per generation. various operators were used in different combinations, and the results analysed. each run generated some 620Mb of data.