DCSCF Detection Method of MMC-DC-Grid Based on Statistical Feature and DBN-SOFTMAX

In the modular multilevel converter-based direct current grid (MMC-DC-Grid), the DC short circuit fault (DCSCF) will bring about the overvoltage and overcurrent in the converter stations. Therefore, the fast and accurate detection of DCSCF will play important role in the application of MMC-DC-Grid. In this paper, a novel fault detection method based on Deep Belief Network (DBN)-Normalized Exponential Function (SOFTMAX) with four statistical feature inputs is proposed. By the proposed method, the features of the DCSCF current, such as the mean, standard deviation, information entropy, and kurtosis values and alike, are firstly extracted and used as the inputs of DBN to complete the training of the entire network through pre-training and backward fine-tuning, and then, the SOFTMAX is utilized to detect the location of DCSCF. The comparisons with several other traditional DCSCF methods, such as Back Propagation (BP), Stacked Autoencoder (SAE), Support Vector Machine (SVM) neural networks, and alike, show that by the proposed method, the DCSCF location can be detected more accurately in a shorter time.