Study On A New Method For Digital Transformer Protection Based On Wavelet Neural Network And FPGA

In the process of building smart grid, which takes Ultra-High Voltage (UHV) as backbone network in our country, with the expansion of power system and the increase of voltage level, power equipments with larger capacity and higher voltage level are needed objectively. More and more large-capacity transformers will be put into operation, which makes stricter and stricter demands of relay protection. The traditional protection methods are under severe challenges. The long time operation experience of differential protection which has been used as the main protection of transformers shows that the transformer'internal and external faults can be distinguished effectively by differential protection. The difficult of differential protection is how to prevent the protection misoperation caused by inrush current.Starting with analyzing the transient mechanism of transformer, this paper respectively founds the mathematical models of inrush current and sympathetic inrush, and researches these inrushes quantitatively by mathematic formulas. On this basis, considering the practical situation, simulation models of inrush current, fault current and sympathetic inrush are built by professional simulation software PSCAD/EMTDC. The generation mechanism, waveform characteristics and effect factor of both inrushes are studied deeply and meticulously.Based on theoretical derivation and simulation research, this paper focuses on inrushes which exist in transformer differential protection, puts forward a new method for digital transformer protection based on wavelet neural network and FPGA:Using db5 wavelet to extract wavelet transform energy characteristic values of inrush current and fault current, and taking these as feature space of improved BP neural network pattern recognition, using the classificatory function of neural network to distinguish inrush current and fault current. In the process of neural network training, this paper adopts the LM optimized algorithm with good robustness and fast convergence. This algorithm can greatly speed up the convergence of network's computing and reduce error of the network training. The simulations of large samples from PSCAD/EMTDC and Matlab software verify that transformer's inrush current and fault current can be distinguished reliably by this method. According to sympathetic inrush, based on a lot of simulations, the conclusion can be drawn that its waveform has no obvious difference with the inrush current's. A comprehensive prevention thought that wavelet neural network methods for identification combined with several preventing misoperation methods is put forward in this paper.According to the characteristics that wavelet neural network's huge computation and high signal sampling rate, a new method that using FPGA which is a high-speed hardware platform is proposed to realize this algorithm. In this paper, several microcomputer protection hardware frameworks based on FPGA are given, feasibility of the algorithm hardware realization plan is proved, which can break the bottleneck that traditional microcomputer protection MCU can not considerate both protection speed and accuracy.In a word, the research results in this paper have high theoretical and practical value in the further improvement of transformer protection.