Investigation Into The Calculation Model And Prediction Method Of Hot-spot Temperature For Oil-immersed Transformers

Power transformers are the main and one of the most expensive parts of electricalnetworks. The operating reliability of large power transformers has a close influence onsecurity and stability of power systems. The end of the life span of power transformersis most due to the loss of their normal insulation, which is very much dependent uponthe highest temperature occurred in any part of a winding insulation system, namely thehot spot temperature(HST). Thus, to optimise the designs and cooling system from athermal point of view, predicting the values and location of HST is essential to meet thegoals of maximizing the load ability, improving the effective lifetime and lowering thetotal cost associated with transformer operation and maintance. Therefore, a lot ofworks focused on calculation the value and location and real time prediction of HST hasbeen developed as follows:①Considering the complicated and special conjugated solid-liquid-gas structure ofoil immersed power transfromer, the inner temperature rise effect arised from heatconduction, convection and radiation is investigated. The mechanism of transformerinternal losses and the influences of no-load losses and load losses on hot spottemperature is also analysised. The temperature rise and drop characteristics of winding,core and oil is given. After the systematic analysis of the heat transfer path, the oil flowand nature convection in vertical duct and horizontal duct is studied and thecorreponding calculating of heat transfer coefficients is given. These theories providedguidance for the establishment of thermal model and numerical simulation for the oilimmeresd transformer.②Based on the heat transfer theory and heat mechanism within transformer, thisdissertation concentrated on establish thermal-electrical analogy model to calculated theHST of transformers. Considering the non-linear thermal resistance, open-circuitimpedance, and the oil viscosity and winding losses with temperature changes are alsotaken into account, proposed an improved model added on the transformer top oiltemperature to calculate HST by using the viscosity and loss correction factors. Modelparameters are estimated by Levenberg-Marquardt method. In the end, by Comparingwith the measured data tested under different conditions, the model shows a goodconsistency, and can describes the temperature variation more accurately in the dynamicloading profiles than IEEE method and ‘swfit’ model. ③For the sake of optimizing the cooling design and predicting the location of HSTfor transformers, numerous study about power transformer thermal actions has beenconducted on a oil-immersed transformer prototype which has axial symmetricgeometry namely, the physical properties of the fluid (i.e. transformer oil) are supposedto temperature function, as well as, in this paper an alternative approach based on FiniteVolume Method (FVM) had been employed to resolve the control equations of flow andheat transfer, which in turn to simulate the overall temperature field and fluiddistribution. The highest temperature disc could be findout based on the lonitudinaldistribution of winding, thus the precise location of HST could be determined by theradial temperature distribution and axial temperature distribution of highest temperaturedisc. It proves this numerical method can not only simulate the fluid distribution withinpower transformer and also predict the value and location of HST effectively.④This dissertation adopted support vector regression (SVR) to establish a modelfor the prediction of HSTs in power transformers. Among which, an improved particleswarm optimization (PSO) with passive congregation algorithm is utilized to determinethe parameters of SVR. The PSO-SVR model has been applied to predict HSTs of apower transformer. Several experimental tests have been carried out involving a reallarge power transformer, to verify the practicality and effectiveness of the proposedPSO-SVR model. In addition, PSO-SVR modeling results are compared with that ofstandard SVR and artificial neural network (ANN) by applying identical training andtest samples. In conclusion, the PSO-SVR model has better prediction accuracy andgeneralization ability than both the standard SVR model and the ANN in the HSTprediction of power transformers.⑤For transformer which has little monitoring information, it is difficult to chooseenough characteristic parameters to establish the SVR model to predict HSTs.So theframework of Kalman filter algorithm was proposed to establish a real-time estimationmodel for HSTs. Results show that the Kalman filter-based model can not only smoothcausal factors and eliminate random noise through the interpolation and filtering torestore the real hot temperature,but also exhibited potential applicability and generalityin real-time prediction for HST,which demonstrates that the proposed model tracingHSTs according to fewer monitoring information and historical data.