Intelligent generation of candidate sets for genetic algorithms in very large search spaces

We have been working on how to select safety measures for space missions in an optimal way. The main limitation on the measures that can be performed is cost. There are often hundreds of possible measures and each measure has an associated cost and an effectiveness that describes its potential to reduce the risk to the mission goals. A computer search of such an enormous search space is not practical if every combination is evaluated. It was therefore decided to use an evolutionary algorithm to improve the efficiency of the search. A simple approach would lead to many sets of solutions which were wildly expensive and so unfeasible. Preselecting candidates which meet the cost goals reduces the problem to a manageable size. Preselection is based on rough set theory since cost goals are usually not rigid. This paper describes the methodology of ensuring every candidate is roughly comparable in cost.