The integration of large scale wind farms has brought new challenges to the transient stability assessment (TSA) problem and difficult measurement of data results in fewer samples. In order to assess the state of the system in the case of small sample data, a deep residual learning (DRL) algorithm that can train deeper neural networks to avoid gradient vanish and gradient explosion is proposed. Firstly, the original input features are constructed by using the data to describe the dynamic characteristics of the power system. Secondly, the DRL trained is applied to the TSA problem. Finally, Compared with the plain convolutional neural networks, the proposed DRL achieves the higher accuracy, moreover, it has the highest unstable F1-score and stable F1-score. Case studies on the modified IEEE New England 39-bus system with wind farms integration exhibit the effectiveness of the proposed algorithm.
Keywords wind farms, transient stability assessment, simple data, deep residual learning, plain network