Cost of uncertainty about controlled objects

The paper aims to quantify, in terms of closed-loop performance, the cost of uncertainty about behaviour of controlled objects. Formulations are introduced to make explicit the dependence of a scalar measure of cost on how the controlled object actually behaves, on what information is available about its behaviour and on how the information is used. A by-product is the definition of two distinct control system properties: ´information variability´ and ´object variability´, characterising aspects of system sensitivity and robustness. General analytic results about the cost of uncertainty are not obtainable for conventional systems, so numerical results are presented for simple first-order LQ systems. A graphical format provides explicit and novel insight into determinants of closed-loop performance. Conclusions are summarised which may be applicable to other more general systems.