Worst-case properties of the uniform distribution and randomized algorithms for robustness analysis (Q1272556): Difference between revisions

As an alternative to the classical worst-case algorithms, a probabilistic approach to robustness analysis and design of uncertain control systems is presented. The aim is to estimate the probability that a system with uncertain parameters restricted to a box attains a fixed level of performance. Two worst-case optimality criteria are defined, and it is proved that both of them the uniform distribution is optimal in the sense that it requires the minimum number of samples to attain a prescribed confidence for all Lipschitz-continuous performance functions. The minimum sample size is calculated which is required to estimate the probability that the volume of the ``bad'' set, essentially the set of parameter values leading to underperformance, is smaller than a certain percentage of the total parameter range volume. The results are applied to uncertain control systems. It is shown how some robustness analysis problems can be put into this setting, and an explicit numerical example is given.