Research On Intelligent Diagnosis Technology Of Transformer Fault Based On The TrueGas

Large power transformer is one of the essential equipments in power system, and its operation state can affect the safety of power system directly. It will cause serious harm to power system and terminal users once there are any faults. Therefore, the fault diagnosis technology of power transformer has important realistic significance.Analysis of fault gas in transformer oil is the effective analysis method of fault diagnosis of transformer. Improved three-ratio method plays an important role in transformer fault diagnosis, but the comprehensive faults can not be detected accurately. The neural network has advantages in the parallel distributed processing, adaptive, associative memory, clustering, fault tolerance and so on. It is fit to judge the happening and development of the transformer inner malfunction and distinguish the multi-malfunction mode. Elman neural network has dynamic information processing capability and its network configuration is simple. But the network convergence speed is slow and convergence accuracy is not ideal. Genetic algorithm has ability of global optimization and improving network convergence speed and accuracy of training. By using the genetic algorithm optimization which optimizes weights and threshold values of Elman neural network, the training speed and precision can be improved, and also the accuracy of fault diagnosis.This paper builds a 5-13-7 neutral network model with the relation between the malfunction gas and the malfunction kind in the transformer. An optimization algorithm, using genetic algorithm to optimize weights and threshold values of neural network, is proposed. Compared with improved three-ratio method and typical Elman neural network, training and diagnosis results show that convergence speed and diagnosis accuracy has improved greatly after the neural network is optimized. After optimized, the accuracy of malfunction diagnosis is nearly 95% and reliability and accuracy of diagnosis is improved greatly, which shows the superiority of genetic algorithm optimization. Software analysis interface of system is programmed by Visual C++. By reading the file, fault gas concentration is real-timely displayed. According to the gas concentration, the value of three-ratio-code is calculated, and system operation is analyzed based on neural network. By calling the database query function, history data can be gotten. With all the operations above, diagnosis is visualization and simplification.