Training sample formation for intelligent recognition of circuit breakers states patterns

This paper presents the system of requirements to the training sample for Intelligent recognition of circuit breakers’ states patterns. To determine optimum parameters of the training sample a series of calculations by means of XGBoost algorithm performed in Python 3 has been carried out. As a result, requirements to size, entropy and informational content of the training sample parameters have been developed. Two oil U- 110-2000 breakers installed on a real 500/220/110 kV substation have been chosen as a calculation example. Requirements to the training sample for a problem of recognition of circuit breakers states patterns have been confirmed. The offered criteria can be used for training of machine learning model as a part of the automated system circuit breakers technical state assessment. Similar system will allow optimizing schedules of power equipment repairs.