Hybridising Self-Organising Maps with Genetic Algorithms.

The aim of this work is to develop a hybridised approach consisting of the Self-Organising Map (SOM), which is an artificial neural network, and a Genetic Algorithm (GA) to tackle large and complex optimisation problems by decomposition. The approach is tested on the travelling salesman problem (TSP), which is a very important problem in combinatorial optimisation. The SOM clusters the nodes into a specified number of groups. Then, the resulting smaller TSPs within each cluster are solved, and finally, the solutions of the clusters are connected to build a high quality solution for the whole TSP. In the two latter steps, a GA is used. Our approach is examined on some random examples including 100 to 10000 nodes as well as 12 benchmark instances. The results with various numbers of clusters are compared to each other and also to the case of solving the TSP with all nodes and without any clustering. Furthermore, the effect of each component algorithm in the hybridised approach is evaluated by some complementary experiments in which the clustering and the optimisation algorithm are replaced by other methods.