Studies On Thermal Circuit Calculation And Genetic Support Vector Machine Prediction For Hot-spot Temperature Of Oil-Immersed Transformer

The power transformer winding hot-spot temperature is one of the main factors ofthe transformer insulation life and the guidance of transformer load operation mode,accurate calculation and prediction of different loads and the hot-spot temperature oftransformer operation in ambient temperatures are of great significance to transformerlife assessment and safe operation.The paper deeply analyzed the relationship of the internal loss of the oil-immersedpower transformer and the reason of internal heat generation, the heat generated by thecore loss directly impact on the top oil temperature rise, basic copper loss and stray lossmainly impact on the hot-spot temperature of heat winding. On this basis, thetransformer heating, heat dissipation and cooling mode was also analyzed，and deducedthe oil-immersed transformer temperature rise curve, laid the foundation for theresearch of hot-spot temperature calculation and prediction methods.Heat transfer theory and thermal-electrical analogy had been adopted, defined thelumped heat capacity and non-linear thermal resistance, and considered the influence ofoil viscosity changes with temperature, tank outer wall heat transfer with thesurrounding environment, winding heat capacity and winding nonlinear heat resistanceimpact on the top oil temperature, a improved dynamic thermal circuit model ofphysical meaning for power transformer was introduced, the model overcomes theinadequate of IEEE Std C57.91and IEC354guide recommend methods. The resultswere compared with temperatures measured data from the laboratory ONANoil-immersed transformer and temperatures calculated by IEEE, the precision of thecalculating results of the model is higher than temperatures calculated by IEEE, namelyit indicates the model have a better agreement with the measured data, verifies that themodel correctness and feasibility.In addition, this paper based on support vector machine can better solve the smallersample learning problems as well as the hot-spot temperature changes over time into anon-linear relationship, and the transformer internal temperature data have thecharacteristics of a small sample, a model of transformer winding hot-spot temperatureprediction model of support vector machine was introduced, it chose genetic algorithmto obtain the optimum parameters of the model. To train and verify of the model basedon oil-immersed transformer temperature rise test, the model was used to predict the winding hot-spot temperature. Studies show that the predicted results are consistent withthe measured values, predict performance and results accuracy is better than BP neuralnetwork and Elman neural network. The result verify the validity of the model and andprovide a new approach and indirect calculation and auxiliary prediction method oftransformer winding hot-spot temperature.