| As one of the critical and numerous electrical equipment in the power system,the transformer whose health status directly associate with the safe and stable operation of the power grid plays the role of transforming and transmitting energy.Transformer hot spot temperature affects the aging of transformer insulation directly,and also determines the transformer can be in a safe and reliable operation state or not.Because of the complexity of transformer structure and the nonlinearity of material parameters,the process of heat generation and dissipation in operation is very complicated.It is difficult to accurately describe the thermal characteristics of transformer by using empirical formula method and thermal circuit model method.However,the existing artificial intelligence algorithms have not been able to establish hot spot prediction models with physical significance,and training samples mostly rely on experiments which limits the universality and extensibility of the prediction model.Therefore,it is necessary to research the prediction model which has physical significance and can obtain the windings hot spot temperature accurately,so as to ensure the safety in the run of transformer equipment.In this paper,S13-M-100 k VA/10 k V oil-immersed self-cooling transformer is taken as the main research object.An inversion detection method of oil-immersed transformer winding hot-spot temperature based on oil flow morphology analysis and support vector regression machine which can achieve a good inversion result under the fluctuation of load coefficient,wind speed and ambient temperature is proposed.And the validity of the method is verified by comparing with the experimental results.The main research work in this paper are as follows:Firstly,it is built a test platform which carried out multi-stage heating and cooling tests of different loads and wind speeds to measure the temperature rise of transformer.The effects of loading time,loading coefficient and wind speed on temperature rise at each temperature measuring point of transformer were analyzed.Based on the actual test model,the finite volume method is used to simulate the temperature field of transformer.The numerical simulation results are compared with the experimental results,which verifies the accuracy of the simulation calculation and lays a foundation for temperature inversion.Then,on the basis of transformer temperature rise measurement platform and temperature rise calculation,combined with support vector regression machine,a streamline method for transformer winding hot spot temperature inversion based on oil flow morphology analysis is proposed.The orthogonal training samples that constructed by the orthogonal test method and the proven temperature rise calculation model are used to invert the steady-state transformer hot spot temperature in different working conditions and environments beyond the training samples.In addition,three characteristic quantities represent the load change are proposed.Some of the experimental datas are selected as training samples to carry out the real-time inversion for the remaining samples’ hot spots.The results verified the advantages of this algorithm in the inversion of transformer hot spots temperature.Finally,it is builded a hot spot inversion model for three-phase unbalance based on the randomness of single-phase load.The maximum load coefficient of the three-phase load is proposed as the load characteristic quantity,and the corresponding temperature measuring points of the shell are selected by observing the mainstream line when different phases are hot spots.At the same time,aiming at the problem of too many shell points are selected when the hot spot locations are irregular,a feature dimension reduction optimization method based on genetic algorithm is proposed.The robustness of model is tested by adding random noise to simulate measurement errors in real situations. |
PDF Full Text Request