| abstract |
The invention relates to a method for detecting modes of a dynamic system comprising a plurality of modes having respectively a set α(t) of characteristic system parameters. According to this method, a chronological series of at least one system variable x(t) is subject to a modelling, such as a switch segmentation, designed to detect a predetermined prediction model, such as a neuronal network relative to a corresponding system mode, in each time interval with a predetermined minimal length for each system variable x(t). After modelling the chronological series, a drift segmentation is performed wherein, a sequence of mixed prediction models is detected in each time interval where the system shifts from a first system mode to a second system mode, said sequence being a result of a linear twin superposition of the prediction models of both system modes. |