Time optimal switching commands by adaptive genetic algorithms

Many of the most effective algorithms used to solve minimum time control problems need a suitable starting input solution. The choice of a correct starling policy is necessary to avoid suboptimal solutions. Since it is not always easy to dispose a suitable initial policy, it is necessary to estimate it numerically. This problem is studied by means of a two phase adaptive genetic algorithm using no prior assumption about the manoeuvre duration, switching number and their time locations. The only information used is the assumption of a bang-bang input command. The proposed adaptation technique consists of the reduction of the control action time by decreasing the chromosomes length, once a solution guaranteeing the defined target error is found. The effectiveness of the procedure was compared with analytical solutions for second and fourth order flexible linear systems.