APPLICATION OF NEURAL NETWORK TO THE DETECTION OF POWER TRANSFORMER RUNNING STATE

A neural network based power transformer running state detection system is devised by using the pair network method to simultaneously process gas chromatogram data and electric experimental data. The system employs fuzzy technique to preprocess input data, uses redundant attributes to improve the learning ability, and utilizes VC dimension to determine network topology. SuperSAB algorithm is adopted to train the network. Experiments and field test of the system show that this system works well in real environment.