Method of automated construction of pattern recognition algorithms on phase paths

Abstract

The problem of automated construction of recognizers of anomalies in the behavior of complicated dynamical systems is solved by means of analyzing trajectories obtained from sensors surrounding the system. A specific feature of the problem consists in the fact that, depending on the individual properties of the system and conditions of its operation, trajectories that contain anomalies may significantly differ from each other in amplitude and length. Besides, the training set could be incompletely defined. The algorithm described here is based on the idea of applying an algebraic approach to the labeling of trajectories. It allows to construct recognizers of abnormal behavior of complicated dynamical systems. The training of the algorithm could be done on an incompletely defined training set.

Keywords

• machine learning
• recognition algorithm
• training set
• problem of constructing an algorithm on a training set
• algebraic approach