Simultaneous optimisation of both scalar parameters and agent reaction strategies using genetic algorithms

When optimising simulation models of organised systems, it may be necessary to optimise the decision making behaviour of human operators and/or computer controllers, either because the optimal strategies for them to use are not self-evident, or because they are dependent on other variables which are being optimised. In this paper, a new kind of genetic algorithm (GA) is presented which optimises problems containing both traditional scalar parameters and multiple reaction strategies, expressed as agent rule bases. It is a multi-level GA in the sense that the strings may have one or more GAs embedded within them. Its performance on two industrial simulation test problems indicate that it can successfully generate good solutions to problems that have relatively small-scale control strategies to be optimised in conjunction with other parameters