Convolutional LSTM-based Frequency Nadir Prediction
- Resource Type
- Conference
- Authors
- Chen, Qingyue; Wang, Xiaoru; Lin, Jintian; Chen, Longyu
- Source
- 2021 4th International Conference on Energy, Electrical and Power Engineering (CEEPE) Energy, Electrical and Power Engineering (CEEPE), 2021 4th International Conference on. :667-672 Apr, 2021
- Subject
- Power, Energy and Industry Applications
Training
Time-frequency analysis
Machine learning
Power system stability
Predictive models
Stability analysis
Loss measurement
deep learning
frequency nadir prediction
power system dynamics
- Language
The frequency nadir of post-disturbance power system is crucial to system stability. To predict this value fast and accurately provides a valuable reference for online security and stability control strategies. As methods based on physical models and shallow learning are getting harder and harder to improve prediction performance, this paper proposes a method for predicting frequency nadir of power system based on convolutional long short-term memory (ConvLSTM) network, fully exploiting temporal and spatial dependencies from measurement data. This approach takes four measured electrical quantities related to frequency dynamics as model inputs and improves the prediction model by adjusting input time series. The experiment of the New England system shows that the proposed model performs better than machine learning methods e-SVR and LSTM. Moreover, it is effective even with incomplete measurement data due to data loss or configuration of measuring devices.