Stationary point of significance level for non-stationary distribution functions

Self-consistent stationary level of non-stationary time series is investigated. The various purposes of practical use of this statistics are discussed. The main problem is an evaluation of significance level of any sample statistics. Unlike the classical problem of stationary test of two samples, in the case under consideration one should construct an indicator to predict the change in the non-stationary regime. For example, we consider special predictor of an attack of epilepsy.