Detection of Abnormal Control Parameters of Thermal Power Unit Based on Modal Decomposition and Deep Learning
摘要
To address the issue of abnormal detection of main control parameters in thermal power unit, this paper proposes a detection method based on Adaptive Variational Mode Decomposition (VMD) and Long Short-Term Memory Variational Autoencoder (LSTM-VAE). Firstly, the main control parameter data is adaptively decomposed using VMD to obtain time series representing dominant modes of the signal. Secondly, LSTM-VAE is used to train the time series of each dominant mode, resulting in a predictive model for the main control parameter data. The reconstruction error loss distribution between the reconstructed values and the actual values is examined to set a reasonable threshold for anomaly detection. Field data experiments demonstrate that this method effectively detects abnormal data samples of main control parameters.