Abnormal Detection of Main Control Parameters of Thermal Power Units Based on VMD-LSTM

In view of the characteristics of weak periodicity of the main control parameter sequence of thermal power units, severe noise interference and the difficulty in identifying tiny anomalies, this study presents an anomaly detection approach for time series data by integrating variational mode decomposition with long short-term memory networks. The input signal is decomposed into multiple modal functions by using variational modal decomposition. The decomposed components are modeled by adopting the long short-term memory network to predict the parameter changes under normal working conditions. Anomalies are detected through the deviation between the actual values and the predicted values. The experimental results demonstrate that this approach achieves remarkable performance in terms of anomaly detection accuracy rate, recall rate, and $\mathbf{F 1}$-score metrics, exhibiting high detection precision.