On convex parameterization of robust control design for minimizing (conditional) performance at risk

This paper introduces performance at risk and conditional performance at risk as design metrics for the formulation of robust control design. These two metrics are used to characterize the high percentile or tail distribution of a performance specification when system uncertain parameters are random variables described by statistical distributions. The probabilistic robust control design is then formulated as a minimization problem with respect to the (conditional) performance at risk or as a constrained problem in terms of them. Performance specifications in terms of the high percentile or tail distribution are more stringent than that are defined in terms of the average (mean) value, which are often used in current literature for probabilistic robust control. Furthermore, the convexity of the conditional performance at risk does not have particular requirements on the underlying distribution of uncertain parameters; thus, convex optimization can be applied to the probabilistic robust control with respect to uncertain parameters with general distributions. The proposed probabilistic robust approach is applied to search solutions to linear matrix inequality containing random parametric uncertainties as well as to design a stabilizing controller for polynomial vector fields subject to random parametric uncertainties.