NEURAL NETWORK PREDICTION OF LEAKAGE CURRENT BASED ON THE THEORY OF TIME SERIES FORECASTING

Abstract

Means of measuring and controlling the value of the leakage current have proven themselves as an effective technical method of monitoring the state of the insulation of the electric motor. The use of devices that allow not only to fix, but also to predict the achievement of dangerous values of the leakage current, which makes it possible to inform the service personnel in advance about the possible danger, and thus reduce the time for simple electrical equipment and use the technological pause for maintenance, repair or replacement of electric motors without waiting their complete rejection. Neural networks used to predict the reliability of electric motors have proven to be effective in predicting these complex processes. On the basis of the obtained experimental data, neural networks were synthesized, both on the basis of technological parameters and on the basis of the theory of time series. A comparison of the operating features of a neural network based on technological parameters and a neural network based on the theory of time series indicates that: the first type of neural network works more efficiently with sharp emissions of the predicted leakage current; the second type of neural networks more accurately models the value of the predicted value near its relatively averaged readings. The peculiarities of the prediction of these neural networks proved the need to create a selection criterion responsible for choosing the most effective of the synthesized neural networks at a certain point in time.