Open Access

ARTICLE

Vibration Trend Prediction of Pumped Storage Unit Based on IBiGRU-KAN and IESC-KELM

Ziwei Zhong¹, Lingkai Zhu¹, Yuheng Zhang¹, Zhiqiang Gong¹, Wei Zheng¹, Wenlong Fu^2,*
1 State Grid Shandong Eectric Power Research Institute, Jinan, China
2 College of Electrical Engineering and New Energy, China Three Gorges University, Yichang, China
* Corresponding Author: Wenlong Fu. Email:
(This article belongs to the Special Issue: Advances in Artificial Intelligence and Machine Learning for Next-Generation Energy Forecasting)

Energy Engineering https://doi.org/10.32604/ee.2026.081825

Received 09 March 2026; Accepted 29 April 2026; Published online 05 June 2026

Abstract

As a key equipment of the power system, the operational stability of pumped storage units (PSUs) is crucial for the safety and efficiency of the power grid. Since vibration serves as a critical indicator of the operational state and structural health of PSUs, accurate vibration trend prediction plays an important role in ensuring the stable operation. To enhance the prediction accuracy of PSUs vibration trend, a novel prediction method is proposed in this paper based on combination of the improved bidirectional gated recurrent unit Kolmogorov-Arnold network (IBiGRU-KAN) and kernel extreme learning machine optimized by the improved escape algorithm (IESC-KELM). Firstly, variational mode decomposition (VMD) is employed to decompose the original vibration signal into multiple subsequences, effectively reducing its nonlinearity and non-stationarity. Secondly, singular spectrum analysis (SSA) is applied to extract the dominant components of each subsequence, while the residual components are superimposed and reconstructed to preserve critical features. Subsequently, training and test sets are constructed using phase space reconstruction (PSR). Then, IESC-KELM is utilized to predict high-frequency subsequences, while IBiGRU-KAN is utilized to predict low-frequency ones. Ultimately, the prediction results of all subsequences are superposed to obtain the final vibration trend prediction result for PSUs. The case analysis and comparative experiments show that compared with the optimal contrastive model, the proposed model has reduced RMSE, MAE, and MAPE by 0.73%, 6.08%, and 6.55%, respectively, verifying the excellent prediction accuracy and stability of the proposed method.

---

Keywords

Pumped storage units; vibration trend prediction; variational mode decomposition; bidirectional gated recurrent unit; kernel extreme learning machine; improved escape algorithm

---