The L-curve in regularisation of optimal control computation

Abstract

We define the L-curve for a regularised solution of an optimal
control problem, and give a method for computing and locating an
optimal regularisation parameter, corresponding to the L-curve's
corner . The regularisation parameter corresponding to this
corner yields a smoother control function, with only a slight change
in the value of the objective. Implementation in the optimal control
software MISER3, and computations of test examples are discussed.