Failure Rate Model And Fault Diagnosis Of Oil-immersed Transformers

Power transformer is core component in power grid and core asset of power companies. Transformer’s safe and stable operation is of great significance. With real-time monitoring of transformer health state and operating conditions, failure rate of transformer can be evaluated, and maintenance measures could be arranged to reduce the risk of equipment failure. In addition, for the transformers with high failure rate and the transformers out of service, fault diagnosis could reveal causes quickly and then remedial measures could be taken to shorten outage time, reducing economic losses caused by transformer outage. Researches on transformer failure rate model and transformer fault diagnosis method are carried out in this dissertation.The Cox’s time-dependent proportional hazard model (PHM) is widely used. Its hazard function contains two parts:the baseline hazard function which represents the aging process, and the link function which represents the equipment status. Based on PHM, a new failure rate model for oil-immersed transformer concerning both aging process and equipment inspection information is proposed. The baseline hazard function follows a temperature-based aging model. The quantity of the dissolved gas is used as the covariate of the link function, for its comprehensiveness. The probability density function of the time to failure is determined and the maximum likelihood estimation (MLE) is used to estimate the model’s parameters. The case study proves the efficiency of the proposed model, which is able to characterize the failure rate of oil-immersed transformer reasonably and comprehensively and can represent the maintenance effect properly.Based on support vector machine (SVM), a new transformer fault diagnosis method which gives fault type probabilities is proposed. An additional sigmoid function is trained to transfer the traditional SVM binary outputs0and1into probabilities of latent faults. With the results of binary classifiers, multi-class probabilities are determined by convex quadratic programming. Providing comprehensive fault type probability information, the proposed fault diagnosis method can achieve a high classification accuracy and still give rational results while other methods misdiagnose. For those samples that might be misclassified using traditional methods such as SVM and three-ratio method, the proposed method can still give the most probable fault type and remind the decision-makers that the real fault type may be hidden in two or three likely types.Laboratory research team develops a software,"Power transmission and transformation equipment condition based smart grid dispatch decision support system." The features of this system are introduced. Finally, the transformer condition assessment, transformer fault diagnosis module and SVG based information display module of the system are presented, which are developed by the dissertation author.