Model Validation from Estimated Closed Loop Performance

摘要: Abstract One reason to make identification experiments is obtain models which can be used design controllers. Structures, for instance model orders determined using statistical tests on the error between output and of system. However, these do not determine how well system controlled. interesting problem predict performance closed loop system, when a controller has been designed based identified model. The purpose this paper show way test sensitivity models. uncertainties parameter estimates are give prediction performance. This done without making any further An algorithm described evaluates increase in expected cost due p~rameter uncertainties. new compared with other numerical results from simulated data given. shows that useful complement validation.