Dissolved Gas Analysis In Power Transformer Oil Using Artificial Neural Network

Power transformer is one of the most important devices in power networks, and it has great effect on the stability and security of power system. Dissolved Gas Analysis (DGA) is an effective method for the detection of incipient faults in transformer. To improve the capability of interpreting the result of DGA, artificial neural network (ANN) is proposed in this thesis. The contributions and conclusions are made as follows:The development of gases dissolved in transformer oil is analyzed, and emphasis is focused on the relationship between the faults and the gases dissolved in transformer oil. An overview and classification of artificial neural networks is presented. The suitable range of conventional ANN models is discussed. Furthermore, the advantage of using artificial neural network in transformer fault diagnosis and prediction is shown.According to the literature and experimental experience, the BP neural network is widely applied in pattern recognition because of its strong ability to approximate nonlinear mappings. Therefore, a combined BP neural network that gives a series steps for power transformer faults diagnosing is provided. The results show the diagnosis accuracy is over 85% by means of the proposed model.The changes of gases dissolved in transformer oil are interrelated. For precise and reliable fault prediction, it is essential to consider several gases simultaneously. A multi-variable predicting model using RBF (Radial Basis Function) neural network is established and apply to dissolved gas analysis. Results show that the proposed model can achieve satisfactory forecasting accuracy.There are various trends in the development of the gases in transformer oil, so it is difficult to describe them accurately by a single model. A combined model which is composed of GM(1,1) model, RBF neural network and BP neural network is presented. The advantages of the combined model over conventional models are shown.Based on the investigations mentioned above, an intelligent power transformer diagnostic and prediction procedure is developed. The effectiveness and advantages of the proposed procedure is shown by the data of a 110 kV power transformer.