文摘
The problem of recognizing segments with abnormal behavior in multidimensional phase trajectories of dynamic systems is considered. Algorithms solving this problem with the help of the axiomatic approach to abnormal behavior recognition in dynamic systems are described. Two parametric families of recognition algorithms based on this approach are considered. For the first parametric family, a necessary condition for the existence of a recognition algorithm making a bounded number of errors on an arbitrary phase trajectory is proved. It is shown that this condition can be violated in the course of learning the first family, but this cannot take place with the second family. Results of a numerical study are presented that demonstrate that the second family provides a better recognition quality and requires less time for training an algorithm than the first family.