An Auxiliary Fault Identification Strategy of Flexible HVDC Grid Based on Convolutional Neural Network With Branch Structures

Abstract

Lightning disturbance may be misjudged as dc fault by the primary protection in the flexible high voltage dc (HVDC) grid. To solve this problem, an auxiliary fault identification strategy based on convolutional neural network with branch structures (BR-CNN) is proposed in this paper. In the proposed scheme, the voltage and current characteristic matrix is constructed as the input matrix of BR-CNN model and the output categories include positive pole-to-ground (PTG) fault and lightning disturbance. Voltage and current branches are constructed to extract high-level local features of input data layer by layer, and main branch is designed to realize the comprehensive utilization of voltage and current information. Through autonomous learning of the model, the nonlinear mapping relationship between input and output is constructed. The method only uses the single-terminal quantities, and can be used as an auxiliary criterion to improve the reliability of the primary protection. The test results verify the effectiveness of the method, and the recognition accuracy is better than the traditional classification models.

Data-Driven Fault Detection and Classification for MTDC Systems by Integrating HCTSA and Softmax Regression

“…Six state-of-the-art AI-based fault identification methods are considered here for the comparison, including five traditional shallow data-driven methods (i.e., ANN with FFT of the DC fault current as the input [20], ANN with Daubechies4 WT of the fault current as the input [21], ANN with the first three principal components of the fault current as the input, SVM with WT entropy as the input [19], and fuzzy-neural patter recognizer [22]), and one deep architecture method (i.e., CNN utilizing the original current signal as the input [24]). The 10-times cross-validation results of the above-mentioned methods are presented in Table VI.…”

“…Recently, several artificial intelligence (AI) algorithms such as Principal Component Analysis [16], Naïve Bayes [17], K-means cluster [18], Support Vector Machine (SVM) [19], and Artificial Neural Network (ANN) [20]- [24] are utilized in DC grid protection. Compared with conventional approaches, AI algorithms make it possible to dig out and synthesize multi-dimensional features from fault transient signals, which may enhance the comprehensive performance of the protection scheme.…”

“…In MMC-MTDC systems, line boundaries are normally weak due to the small inductance of DC reactors and equivalent capacitance of converts, and thus more careful feature selection is in demand to capture the subtle differences between some internal and external faults. By contrast, the latter class [24] embeds the feature extraction into the deep neural network, so that the original transient waveforms can be directly inputted towards the data model, and useful features can be automatedly determined in the training process. In [24], a branch-structure convolutional neural network (CNN) is designed for fault identification in MMC-MTDC systems.…”

“…By contrast, the latter class [24] embeds the feature extraction into the deep neural network, so that the original transient waveforms can be directly inputted towards the data model, and useful features can be automatedly determined in the training process. In [24], a branch-structure convolutional neural network (CNN) is designed for fault identification in MMC-MTDC systems.…”

Multi‐vendor interoperability in HVDC grid protection: State‐of‐the‐art and challenges ahead

“…Additionally, a combination of different criteria may be adopted to improve the selectivity of a protection scheme [61,64]. In recent years, data-driven approaches using advanced machine learning algorithms are seen attracting attention in an academic context, such as support vector machine [65] and convolution neutral network [66]. The fault detection speed of local measurement-based algorithms is often very fast and proportional to fault distance.…”