Abstract: With the enhanced trend of alternative clean energy and power electronics in power system, the traditional numerical simulation methods based on theoretical model will face new challenges, while the self-adaptive data-driven power system stability assessment method is gaining more and more attention. Based on the theory of extreme learning machine (ELM), a transient stability assessment method suitable for online application is proposed. Firstly, the samples are screened by adjusting the ratio of stable and unstable simulation samples, to reduce the sample imbalance in which unstable samples are far less than stable ones in the sample set, and the recursive feature elimination is used to further process the sample set. Then, the cross-validation is used to optimize the ELM network structure, and the processed sample set is used to train the ELM. Finally, the system stability based on the output of the neural network is predicted, and the reliability of the results with the improved evaluation criteria is evaluated. Test results show that the recursive feature elimination can significantly reduce the feature redundancy and improve the performance of the model, and the proposed algorithm has a shorter training time while can provide more accurate results.

Key words: power system; transient stability; extreme learning machine (ELM); recursive feature elimination; cross-validation