基于油中溶解气体分析的变压器内部故障诊断方法分析-analysis of transformer internal fault diagnosis method based on dissolved gas analysis in oil.docxVIP

AbstractThe article begins with the introduction of the typical transformer fault, studying the principle of the gas production of the gases dissolved in oil and the corresponding relation between the typical transformer fault and the gases dissolved in oil. On the bases of a great amount of collection of fault characteristic, we set up a sample book of transformer fault to provide supporting data for neural network.However, the relationship between the characteristic and fault is relatively complex, uncertain and non-linear, so it is difficult to obtain the mathematical model for traditional fault diagnosis. Because of the distinct advantage of ANN, such as adaptability, Self-study, distributed parallel processes, association memory and the nonlinear reflection ability, it gives a new approach to solve above-mentioned problems.The thesis simulate in Matlab using a variety of algorithms and the simulation result proves that, BP network gets local minimization too easily with the enlargement of fault samples in scale witch will brings great challenge to the accurate diagnosis.To overcome the shortcoming of BP network, this paper present a predicting model of transformer fault diagnosis based on PNN witch has good effects in pattern recognition. The advantages of PNN lie in its linear learning algorithms instead of nonlinear algorithm, moreover, keep a high-precision characteristic of nonlinear learning algorithm. This weight value of the network is the distribution of samples, so the network is not needed in training that can completely meet the requirements of real-time processing. Simulation results show that transformer fault diagnosis using PNN greatly increase the diagnosis speed and precision, witch is better than BP network in diagnosis speed,additional sample and fault tolerance features. To improve further degree of accuracy, the thesis provides a PNN based on δ optimization using the genetic algorithm.Simulation results show that this method can effectively