Oil-immersed Power Transformer Fault Diagnosis

The rapid development of society economy promotes the ongoing growth of total electricity consumption, and grid systems work in the long-term overload state, which has brought many problems for the maintenance personnel. Power transformer is one of the key equipment, then the safe operation of power transformer has gained more and more attention, therefore the potential fault detection of power transformer has became an hard research topic. Internal faults of power transformers includes electrical failures and thermal failures. Once the occurrence of faults may cause the system can't safe and normal operation, and even serious accidents. So the timing diagnosis of power transformers is significance.For fault diagnosis of oil-immersed power transformers, this paper proposes two solutions. The first scheme is proposed for fault diagnosis accuracy. Due to the border demarcation unclear of the IEC three ratio method and the different samples of types of faults in the oil dissolved gas data samples,results in the unbalanced samples, and these factors lead to low diagnosis accuracy. Therefore the paper has proposed a fault detection scheme of power transformers based on GA to optimize the rule table. According to characteristics and types of transformer faults, established fuzzy IEC three ratio fault diagnosis system and used the bootstrapping expansion DGA samples. In order to the higher diagnosis accuracy, this paper also adopted GA to optimize fuzzy rule table.The second method is for BP network easily into the local minimum and "premature" issues of genetic algorithm(GA). The paper mainly used simulated annealing algorithm(SAA) to optimize wavelet neural network(WNN). The advantages of SAA is to avoid falling into local minimum, while the wavelet analysis of WNN has good properties in time domain and frequency domain. Which make up the defects of SAA in convergence rate.Finally, the paper verified these two programs with the Matlab simulation software. Through the comparison and analysis of the simulation results, the first program have a higher diagnostic accuracy. While the simulation results of the second way has more advantages than other methods in the problem of falling into local minima. This shows that the two ways is feasible and effective.