Finding multiple solutions with an evolutionary algorithm

A new multiple-solution technique is presented that addresses some of the limitations of existing speciation and multiple-solution techniques. This genetic-algorithm (GA) technique packs multiple problem points within a genotype and a uses a fitness function based on intersolution distance and individual solution fitness. The technique is demonstrated on a contrived multimodal TSP test problem and it is found effective in finding two maximally distant and near-optimal solutions. The technique can be used with a generational or steady-state GA model and does not depend on the explicit use of crossover or a binary-based encoding. Therefore the technique may be of interest in other population-based computational models other than genetic algorithms