Multi-task Transient Stability Assessment Based on Feature-level Fusion of Heterogeneous Data

Abstract

Considering that the existing deep learning-based transient stability assessment methods do not fully describe the input data of power systems, the heterogeneous data is often ignored, and lots of feature information is unable to be effectively fused. In order to make full use of the various kinds of heterogeneous data of the power system to improve the accuracy of the model and the performance of the algorithm, a deep learning method based on the feature-level fusion of heterogeneous data is proposed. Firstly, the multi-layer perceptron (MTL), graph convolutional network (GCN), and gated recurrent unit (GRU) are used to extract the features from static multivariable data, topological graph data, and time series multivariable data, respectively. Secondly, the method of tensor fusion is used to carry out the feature-level fusion of the extracted features. After the flattened fusion features are input into the shared layer, the transient stability discrimination and margin prediction are realized simultaneously by using the multi-task learning method based on the homoscedastic uncertainty. On this basis, a transient stability assessment model is established to evaluate the performance of the proposed method. Finally, the New England 10-machine 39-bus system is adopted for simulation, training and verification. The results show that the proposed method can effectively improve the assessment accuracy and robustness.