Машинное обучение как инструмент повышения эффективности управления жизненным циклом высоковольтного электрооборудования

Abstract

The purpose of the study is to analyze the practical implementation of high-voltage electrical equipment technical state estimation subsystems as a part of solving the lifecycle management problem based on machine learning methods and taking into account the effect of the adjacent power system operation modes. To deal with the problem of power equipment technical state analysis, i.e. power equipment state pattern recognition, XGBoost based on gradient boosting decision tree algorithm is used. Its main advantages are the ability to process gapped data and efficient operation with tabular data for solving classification and regression problems. The author suggests the formation procedure of correct and sufficient initial database for high-voltage equipment state pattern recognition based on its technical diagnostic data and the algorithm for training and testing sets creation in order to improve the identification accuracy of power equipment actual state. The description and justification of the machine learning method and corresponding error metrics are also provided. Based on the actual states of power transformers and circuit breakers the sets of technical diagnostic parameters that have the greatest impact on the accuracy of state identification are formed. The effectiveness of using power systems operation parameters as additional features is also confirmed. It is determined that the consideration of operation parameters obtained by calculation as a part of the training set for high-voltage equipment technical state identification makes it possible to improve the tuning accuracy. The developed structure and approaches to power equipment technical state analysis supplemented by power system operation mode data and diagnostic results provide an information link between the tasks of technological and dispatch control. This allows us to consider the task of power system operation mode planning from the standpoint of power equipment technical state and identify the priorities in repair and maintenance to eliminate power network “bottlenecks”.