Subsynchronous Oscillation Data Generation Method of Power System Based on Bidirectional Long Shortterm Memory Generative Adversarial Network
The subsynchronous oscillation data generation method based on bidirectional long short term memory generative adversarial networks(BiLSTM-GAN)is proposed to address the lack of subsynchronous oscillation data in practical engineering.Firstly,the bidirectional long short-term memory(BiLSTM)network is introduced into the generative and discriminative models to fully exploit the forward and backward temporal sequence information of oscillation data.Then,Wasserstein distance is incorporated into the generative adversarial network(GAN)model to solve the problem of training instability.Finally,dynamic time warping(DTW)-based similarity metrics and frequency domain analysis-based authenticity metrics are proposed to assess the quality of generated samples.Case studies demonstrate that the data generated by the proposed method aligns with the characteristics of oscillation data,exhibits certain advantages in data authenticity.
long short term memory networkssubsynchronous oscillationdata generationWasserstein distance
NETL
NSTL
万方数据
