Narrowing the scope of hidden nodes based on empirical formula to find the optimal number of hidden nodes in the small range is described

aiming at problems of easy trapping into smallest spot. According to the glowworm swarm optimized（ GSO） features

using GSO optimize training initial weights and thresholds of back propagation neural network（ BP-NN） can predict test set well

so as to avoid the problem of BP-NN falling into local minima. Using a combination of the two methods for grid fault diagnosis

experimental results show that the method can accurately and efficiently diagnose network fault location.