A Power System Disturbance Classification Method Based on Enhanced Denoising Autoencoder and Random Forest

To avoid the occurrence of large-scale power outage, the realtime and accurate power system disturbance classification is a helpful measure. However, unsatisfying data quality of synchronous phasor measurement units seriously affects its application in disturbance classification. For this reason, a power system disturbance classification method based on the combination of adaptively weighted long short-term denoising autoencoder and random forest was proposed. Firstly, the long short-term memory was utilized to construct a kind of enhanced denoising autoencoder to establish mapping relation between bad data and normal data was constructed. Secondly, according to the change trend of verification loss of different measurements, an adaptive weighted multitask denoising network, which could adaptively update the weight of loss function corresponding to each measurement to reduce the reconstruction error, was presented. Finally, the random forest classifier was used to classify the encoding features, and the Bayesian optimization algorithm was used to optimize the hyperparameters. The simulation results on IEEE 39 bus system show that the proposed method possesses better accuracy and real-time performance for bad data level from different phase measurement units, and it is verified by field data that the proposed method has higher generalization.