Fault diagnosis of electric power system transformer on CMAC neural network approach

Regarding characteristics of electric power system transformer faults, such as diversity of types, uncertainty of fault information and irregularity, this study employed the Cerebellar Model Articulation controller(CMAC) neural network approach to diagnose dissolved gas in transformers, in order to determine the fault causes. By using the CMAC neural network's memory space, weight memory and other learning mechanisms, similar input data are excited to the same memory locations to achieve the fast convergence results. Good recognition capabilities are available in the recognition of similar data. Moreover, regarding the non-training data, they are categorized according to specific category similarity. This study conducted the five types of electric power system transformer faults for the final fault diagnosis. The simulation results showed that the correct judgment rate reached nearly 94% through a small number of iterations of training times, thus improving the diagnosis accuracy and efficiency.