A gradient method for computing optimal bang-bang controls
H. Sirisena · International Journal of Control · 1974
An established technique for computing optimal bang-bang controls is to minimize the coat of the bang-bang control with respect to the switching times and final time using a direct search method. In this paper, explicit expressions are obtained for the gradient of the cost of a bang-bang control with respect to these parameters. This enables the minimization to be performed more accurately, and also more rapidly, using a gradient search method such as that of Davidon. The use of this approach to compute time-optimal controls for a fifth-order system, and to minimize integral-square-error for a third-order system is described.