Pumped storage power stations face significant efficiency and safety challenges due to complex hydraulic and mechanical faults. While accurate fault diagnosis is critical for reliable operation, existing data-driven methods often lack generalizability and explicit integration of domain knowledge in multi-fault scenarios. To address this, we develop a novel simulation-knowledge hybrid framework that integrates physics-based representations from Computational Fluid Dynamics (CFD) / Finite Element Method (FEM) simulations with a data-driven Stacked Denoising Auto-Encoder (SDAE) model enhanced by RIME optimization algorithm. The framework introduces a systematic knowledge-formalization and feature-alignment mechanism that translates implicit physical fields into quantifiable indicators aligned with experimental features. Through detailed hydraulic and mechanical simulations, we characterize cavitation and bearing wear fault signatures, formalizing them into traceable diagnostic datasets. This establishes a transparent evidence chain linking fault diagnoses to underlying physical mechanisms. To improve generalization, the SDAE model undergoes autonomous hyperparameter adaptation via the RIME algorithm, enhancing its capability to interpret hybrid knowledge-data inputs. Experimental validation demonstrates that the knowledge-integrated RIME-SDAE model achieves near-perfect identification accuracy exceeding 99%, outperforming both baseline SDAE (93.3%) and SVM models. Field tests confirm the framework's robustness and accuracy, enabling real-time fault traceability without additional sensors. This research provides a scalable methodology for enhancing operational reliability and supporting design decisions in pumped storage power stations through explicit knowledge utilization and autonomous model adaptation.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号