Studies On The Prediction Method Of Transformer Winding Hot Spot Temperature Based On BPNN

Power Transformer is one of the most important and critical equipments of the power net, thus to ensure its reliability is significant to the safety and relialibity of the whole power grid. For most transformers the reason for end of life is the loss of due insulation, while one of the main factors to affect the insulation capacity is the winding temperature of service transformers. The transformer winding temperature of the hottest areas, which is called winding hot spot temperature, not only affects the load capacity and winding insulation life, but is an important indicator to evaluate transformer winding design. Therefore, monitoring transformer winding hot spot temperature is important to estimate the transformer life, ensure a safe and reliable operation of the system and improve the economic benefits of operation.The paper, based on the domestic and foreign researches on hot spot temperature, analyzes the heating and cooling mechanism of oil-immersed transformer and sets up a temperature test platform to study the temperature distribution inside the transformer; then according to parameters obtained from the test, the author uses artificial neural network to predict the transformer winding hot spot temperature, and the main tasks are as follows:â‘ Through the study on the process well as the mechanism of the transformer internal heating we find the main factors to affect temperature distribution inside the transformer: environmental conditions, load current and the oil cooling patterns. Accordingly, we deduce the heating and cooling characteristic equation of the main components like winding, core, transformer oil, and get their heating and cooling curves.â‘¡With the combination of the transformer temperature distribution and the practical requirements for winding temperature measurement inside the transformer, we build a transformer temperature test platform on oil-immersed transformer to study the temperature inside the transformer under the form of natural oil circulation natural air cooling (ONAN). The results show the winding temperature distribution along the height is nonlinear, and the transformer winding hot spot temperature happens in the direction along the winding height of about 5/6.â‘¢Based on systematic analysis of transformer internal heating process and temperature distribution, combined with the thermal-electric analogy method and hot spot temperature rise model recommended by IEEE, we propose the transformer winding hot spot temperature calculation simulation model based on thermal-electric analogy. Then the predictive values of the simulation model, IEEE model values and actual measured values are compared, and the results show that under less load and rated load both transformer top oil temperature and winding hot spot temperature are closer to the measured temperature than the IEEE model values, and under over load and variable load all the three values are able to better predict the temperature well.â‘£Using BP neural network to carry out power transformer winding hot spot temperature prediction model, we focus on issues as its network selection, study samples establishment, input selection, and algorithm improvement etc., and establish the improved BP neural network based on the Levenberg-Marquardt algorithm. Simulation results show that the improved three-layer BP neural network can implement transformer winding hot spot temperature prediction well.