基于有限初始采样的直流微电网故障定位广义神经网络方法

High discharge rate of the DC link capacitor during a line fault in a DC microgrid causes a steep rise in the line fault current. Previous line fault detection techniques typically detect a low resistance fault in DC microgrids within a few microseconds; therefore, a fault location estimation scheme using a few initial samples is necessary. This work proposes a local-measurement-based offline fault location estimation scheme using a few initial samples. The proposed generalized approach aims to obtain an optimal Neural Network (NN) to predict the fault location in DC microgrids for each input data set. The proposed scheme is validated for initial samples ($=2$ and 4), making the location estimation time $120\mu$s and $240\mu$s, respectively, with $60\mu$ s sampling time. The estimation accuracy is further refined by averaging each prediction as the ultimate fault location. The proposed scheme is validated using the Real-Time Digital Simulator on a 400 V DC microgrid with different operating modes. Parallel assessment with previous techniques and comparison with traditional NN shows the superiority of the proposed scheme in fault location estimation with a few initial samples.