Accurate fault classification for distribution network can enhance the pertinence of repair

which is of significance to the rapid restoration of power supply. We proposed a fault classification technique for neutral non-effectively grounded system using the binary-ant-colony-algorithm（BACA）-based fuzzy neural network（FNN）. The high-frequency component of the 2 ms’ transient signal after the failure was extracted by using empirical mode decomposition（EMD）. Based on the method of mathematical statistics

the feature vectors of the inference system were structured. Then the change regulation of the feature vectors was researched under different fault conditions. BACA was used to optimize the weights of the FNN

which overcame the shortcomings of slow scouting speed and local minimum. A computational model was structured in ATP-EMTP based on the physical truth

then the inference system was trained with the computing results. The test samples and the results of field experiments showed that the fault classification method proposed in this paper had a high accuracy and good adaptability.