Research On Transformer Winding Temperature Prediction Based On Finite Element Method And FC

As a crucial primary device in power systems,transformers play a pivotal role in ensuring the safe and stable operation of said systems.Among the various accidents that can occur,those resulting from insulation aging due to overheating of the winding and subsequent breakdown account for a significant proportion.Such incidents pose the greatest threat to the safety and stability of power systems and thus,accurately and promptly detecting the hot spot temperature of the winding holds immense importance in terms of improving economic efficiency and extending the lifespan of transformers.It is therefore one of the key areas of focus for transformer manufacturers and research institutions in the power sector.The oil-immersed transformer,due to its superior heat dissipation performance,safe and reliable characteristics,and large capacity advantages,has been widely used in the power system.The main research content and achievements obtained in this article are as follows:(1)Researching a convergence calculation method for transformer fluid-structure coupling based on finite element analysis.Addressing the problems of low accuracy when using the finite volume method to solve complex structures and the difficulty in convergence when using the traditional finite element method to solve fluid fields.The research object is the SZ20-8000 kVA/35 kV strong current circulating cooling transformer structure.Three-dimensional and two-dimensional axisymmetric transformer fluid-solid coupling calculation models were established,and the internal temperature distribution of the transformer under normal conditions and magnetic leakage faults was simulated and calculated.The calculation accuracy of several mesh divisions was compared,and the effects of inlet flow velocity,load,and environmental temperature on the temperature of the transformer windings were analyzed.The study found that the finite element iterative algorithm combining parallel adjustment of boundary conditions and inheriting new convergence solutions as initial values can solve the non-convergence problem well.By using a hybrid mesh division,the calculation time can be greatly reduced while ensuring accuracy.The hot spot temperature of the transformer windings under the two conditions decreases as the inlet flow velocity increases,and the decrease gradually becomes smaller.The higher the load and environmental temperature,the higher the temperature of the winding hot spot.Under magnetic leakage,the overall temperature of the winding will increase by about1 K.When the transformer is in a magnetic leakage state,the winding needs a higher load to dominate the temperature field of the transformer,which will cause a decrease in the working efficiency of the transformer under the same load,resulting in the waste of power resources.(2)Conducting research on a hot spot temperature prediction method for transformers based on fully connected neural networks.The problem of low accuracy and poor model generalization ability in predicting the hot spot temperature of transformers using SVM models needs to be improved.Using simulated data to train nine deep learning models,six of which are FCN models for three-dimensional transformers under normal and magnetic leakage states,affected by inlet flow rate,load,and environmental temperature,while the remaining three are FCN models for two-dimensional axisymmetric transformers under normal state,affected by inlet flow rate,load,and environmental temperature.These models aim to learn the relationship between the transformer temperature field state and internal state parameters to predict the temperature field changes of the transformer.Six nodes were randomly selected from the simulation data to study the effects of inlet flow rate,load,and ambient temperature changes on the transformer temperature field,in order to obtain more accurate results.At the same time,the average temperature of all nodes under each influencing factor was taken to observe the average influence of each influencing factor on the transformer temperature field.The transformer temperature field calculation model based on deep learning can predict the hot spot temperature of the transformer winding,with a prediction error of less than 1%,proving the effectiveness of the model.(3)Exploring a non-invasive and rapid method for obtaining transformer winding temperatures by using cubic spline interpolation to replace the line boundary conditions.In response to the problem of non-invasive acquisition of hot spot temperature of transformer winding based on minimal sensors and simulation calculations,this article takes a natural convection cooling transformer with a structure of SSZ20-63000 kVA/110 kV as the research object and proposes a method for inverting the hot spot temperature of the transformer winding based on cubic spline interpolation to obtain line boundary conditions combined with fully connected neural network.The method extracts characteristic points with large temperature changes on the line boundary and uses the temperature on these points as the initial value of the interpolation function.The interpolated line boundary replaces the original line boundary to calculate the temperature field of the transformer.The computational results of linear interpolation,piecewise cubic Hermite interpolation,and cubic spline interpolation are compared and analyzed.The results show that the cubic spline interpolation has the best effect,with a maximum error of 0.185% compared to the original boundary results and an error of0.675% compared to experimental values.After integrating the FCN model,with good performance in prediction,we trained two models,namely the transformer temperature field prediction model under the third-order spline interpolation boundary method and the heat flux boundary method condition,respectively.The prediction errors for these two models were 0.657% and 0.752%,respectively,which verifies the effectiveness of the inversion method.This method provides practical value in predicting the hot spot temperature of the transformer using a limited number of points with high precision.