LQG auto-tuning

A well known and accepted problem in optimal LQG (linear quadratic Gaussian) controller design is the choice of the quadratic cost function weights to produce desired or acceptable closed-loop characteristics. In this paper two simple methods of automatically designing (auto-tuning) the cost-function weights are introduced and discussed for a specific formulation of the discrete LQG control problem. The first method is only applicable to the control of simple systems where the most complex discrete process allowed is second order and the dynamic weights do not contain an integrator. The second method to be described is a simulation based method which determines a control weight from the simulated closed-loop response to a set-point change. Two performance objectives are investigated and these are given a physical interpretation therefore relating the control weight to a specific type of output performance