WGAN-GP enhanced ITOC-VMD-FBiLSTM for vibration trend prediction in pumped storage units

Pumped storage units (PSUs) are vital components of the power system, and their operational stability is crucial for ensuring grid reliability. Considering that most faults and abnormal operating conditions in PSUs are typically manifested through vibration signals, the accurate prediction of vibration trends plays a crucial role in maintaining equipment stability and enhancing operational reliability. However, traditional models often struggle to capture subtle yet critical fluctuations in the vibration data, which poses significant challenges for reliable forecasting. To address this issue, a novel vibration trend prediction approach is proposed, integrating a wasserstein generative adversarial network with gradient penalty (WGAN-GP) and a Fourier-based bidirectional long short-term memory network (FBiLSTM). First, WGAN-GP is employed to generate synthetic vibration data, which is then combined with the original dataset to enhance the diversity of the data. Next, the parameters of variational mode decomposition (VMD) are optimized using an improved tornado optimizer with Coriolis force (ITOC), enabling accurate signal decomposition, noise reduction, and the extraction of key vibration components. Then, the training and test sets are constructed based on phase space reconstruction (PSR). Finally, the FBiLSTM model is employed to predict the vibration trend, and the prediction results of each decomposed subsequence are superimposed to obtain the final prediction result. Numerical experiments and comparative analysis demonstrate that the proposed method significantly improves prediction accuracy and stability, exhibiting enhanced performance in capturing complex vibration characteristics.