Research on an Intelligent Diagnosis and Decision Support System for Pumped Storage Units Based on Multi-Source Data Fusion and Hybrid Intelligent Algorithms

Abstract

Pumped storage hydropower (PSH) is a key regulating resource for renewable energy integration and power system stability. Due to frequent start-stop operations, deep peak-load regulation, and bidirectional operating conditions, stator winding insulation degradation, rotor inter-turn short circuits, and end-winding vibration have become the dominant failure modes of pumped storage units. Conventional monitoring systems are limited by single-source sensing, asynchronous data acquisition, high misdiagnosis rates, and maintenance decisions that rely heavily on expert experience, making traditional periodic maintenance increasingly inadequate. To address these challenges, this study proposes an intelligent diagnosis and decision support system based on multi-source data fusion and hybrid intelligent algorithms. An Intelligent Electronic Device (IED)-based condition monitoring platform is developed by integrating multiple sensing technologies. Complete Variational Mode Decomposition (CVMD) and Kernel Principal Component Analysis (KPCA) are employed to extract representative features from multi-physical-field data, while an attention-enhanced Long Short-Term Memory (LSTM) network is introduced for accurate fault identification. In addition, adaptive time-alignment and joint denoising algorithms are developed to improve data quality and diagnostic robustness. A predictive maintenance framework incorporating health assessment and remaining useful life prediction is further established to optimize maintenance scheduling. Results demonstrate that the proposed system achieves a fault prediction accuracy of over 90% and reduces annual maintenance costs by approximately 15–20%. The proposed framework provides an effective solution for intelligent operation and maintenance of modern pumped storage units.