MULTI-STAGE METHOD FOR SHORT-TERM FORECASTING OF TEMPERATURE CHANGES MODES IN THE POWER CABLE

Abstract

The article is devoted to research on the creation of diagnostics and prediction of thermofluctuation processes of insulating materials of power cable lines (PCL) of electric power systems based on such methods of artificial intelligence as neural networks and fuzzy logic. The necessity of developing a better methodology for the analysis of thermal conditions in PCL is shown. The urgency of the task of creating neural networks (NS) for assessing the throughput, calculating and predicting the temperature of PCL conductors in real time based on the data of the temperature monitoring system, taking into account changes in the current load of the line and the external conditions of the heat sink, is substantiated. Based on the main criteria, traditional and neural network algorithms for forecasting are compared, and the advantage of NS methods is shown. The classification of NS methods and models for predicting the temperature conditions of cosmic rays has been carried out. The proposed neural network algorithm for predicting the characteristics of electrical isolation was tested on a control sample of experimental data on which training of an artificial neural network was not carried out. The forecast results showed the effectiveness of the selected model. To solve the problem of PCL resource prediction, a network was selected with direct data distribution and back propagation of the error, because Networks of thistype, together with the activation function in the form of a hyperbolic tangent, are to some extent a universal structure for many problems of approximation, approximation, and forecasting. A neural network was developed to determine the temperature regime of a current-carrying core of a power cable. A comparative analysis of the experimental and calculated characteristics of the temperature distributions was carried out, while various load modes and the functions of changing the cable current were investigated. When analyzing the data, it was determined that the maximum deviation of the data received from the neural network from the data of the training sample was less than 2.2 %, which is an acceptable result. The model can be used in devices and systems for continuous diagnosis of power cables by temperature conditions.