A new approach for partial discharge recognition on transformers on-site by means of genetic algorithms

In this paper, the application of genetic algorithms to partial discharge (PD) recognition on transformers onsite is introduced. First, a short description of genetic algorithms is given. Afterwards, the separation of PD from noise signals, as one of the main problems for a sensitive PD measurement onsite, is explained. It is shown that the use of genetic algorithms for the optimization of neural networks allows a sufficient separation of PD from the pulse shaped noises. Furthermore, the use of genetic algorithms for localizing PD origins in transformers is of great promise, if reference signals are available. Here, the application of genetic algorithms is demonstrated on signals measured at the coil of a distribution transformer in the laboratory. The limits of the used methods are shown by superposing synthetically generated noises to the measured pulses