Study On Insulation Fault Diagnosis And Thermal State Parameters Prediction Model Of Power Transformers

Intelligent fault diagnosis technology of power transformer is the premise of correct assessment of insulation state, and accurate grasp of internal heat state of power transformer determines the load capacity and service life in a large extent which also is the extension and expansion of insulation fault diagnosis. Meanwhile, only after correctly assessing insulation state of power transformer, tracking the service life, we can make the right maintenance decisions based on the actual state of power transformer. As a whole view of intelligent fault diagnosis and internal thermal state prediction of power transformer, this paper had an in-depth study on these two vital areas, and the main research works of this dissertation are shown as follows:①Usually, a neural network method based on BP algorithm was employed to solve this problem, meanwhile, we mainly focused on network selecting, study samples building, input modes choosing, convergence of training process and algorithm modifying etc. Finally, this dissertation proposed a neural network based on modified BP algorithm, i.e. momentum coefficient and alter-learning coefficient, then analysis and simulation results show that three-layers neural network based on modified algorithm can build clear mapping between gases-dissolved-in-oil and fault types of power transformers.②This dissertation presented a novel insulation faults diagnosis model of power transformers based on expanded multi-input and multi-output wavelet neural network. After analyzing the diagnostic results differences and its mathematic mechanism, we can draw some conclusions that the proposing wavelet neural networks approaches prevail the traditional BP neural network on convergence speed, classifying ability and diagnostic accuracy.③Combining genetic algorithm and wavelet neural network, this dissertation set up an effective fault diagnostic method of power transformers, i.e. genetic algorithm evolving wavelet neural network, which can reflect the nonlinear relationship between gases-dissolved-in-oil and fault types of power transformers. What's more, employing genetic algorithm is order to let wavelet neural network deal with all information obtained better and classify the faults types more accurately. A number of examples show that the method proposed also has good classifying capability for single-fault and multiple-fault samples of power transformers as well as the highest faults diagnostic accuracy.④After the systematic analysis of interior heating process and temperature distribution of power transformers, this dissertation then placed the focus on the steady state and transient state temperature prediction models. Based on the fundamental heat transfer theory and interior heating process of power transformers, and drawed an analogy between the heat transmission theory and Kirchhoff's law, this dissertation presents a thermal-electric analogous dynamic top oil temperature prediction thermal model of an oil-immersed power transformers which cooling model is OF. A lot of experimental waveforms and analysis show that the proposed model is able to yield better results than IEEE thermal model on predicting the top oil temperature of power transformers in OF cooling model.⑤Considering all kinds of loading status, cooling mode etc, then fully regarding nonlinear thermal resistance and oil viscosity based on interior heating process and temperature distribution, this dissertation built a dynamic hot spot temperature model for OF cooling model power transformers. Plenty of experimental curves proved the predicting temperature can totally reflect the prediction thermal characters of oil-immersed power transformers in OF cooling model, on the other hand, the errors between the measurable values and prediction results are fully within the scope of the permit.