Time-varying Outage Model And Fault Diagnosis Of Oil Immersed Transformers Based On DGA

Oil-immersed transformer is one of the most important equipments in power system, which transforms voltages and transmits power energy. The healthy, safety and stable operation of transformer can lower the operation risk of power system dramatically. Thus, real-time operating condition of the transformers should be known to avoid outages because of some latent factors, as well as lower operating risks. Meanwhile, fault diagnosis of the transformers out of service helps shorten outage time and operating cost. Some key issues are being studied in this paper.Dissolved gases come into being in the insulating oil during the operation of power transformers. By experience, it is known that the healthy state of transformers can be deduced from the content of the dissolved gasses and the development of the internal latent fault in that state can be estimated by the generating rate of the gasses. Based on this knowledge, the time-varying outage model for risk assessment of transformers is built. The transformer is firstly decided by the quantity of the dissolved gas. Thus the multi-state Markov model is set up considering the maintenances to get the basic failure rate. Meanwhile, the time-varying exponential outage model is established based on the gas generating rate which reflects the operating characteristics as well as the development of the latent fault. The model is validated by case studies and provides reference foundation for the device risk assessment.Support Vector Machine (SVM) does well in fault diagnosis of power transformers which is a small sample classification problem. It is sensitive to the kernel function, the kernel parameter and the model parameters, which could be lowered by multiple kernels learning. This paper presents a novel method named BPSO-MKSVC for fault diagnosis of power transformers. Multi-kernel support vector classifier (MKSVC) uses a combined kernel formed through a linear combination of several basis kernels, each of which extracts a specific type of information from the training data, providing a partial description of the data. Given many partial descriptions of the data, a convex optimization is obtained by a linear combination. The binary particle swarm optimization (BPSO) method is adopted to help samples choosing appropriate basis kernels. Case studies show that the proposed model can achieve better diagnosis results with lower parameters sensitivity and better robustness.Based on the algorithm analysis above and the IDP 90 platform, the electric transmission and transformation equipment condition management software, mainly the fault diagnosis of transformers module, is developed. The practical use of the methods is realized through the module and the use of the software is shown up by operation.