Fault Diagnosis of Hydro-Power Units Using BP Neural Network and XGBoost Algorithm for Enhanced Operational Safety

To enhance operational safety and reduce maintenance costs, this study investigates the fault diagnosis of hydro-power units, where the BP neural network and XGBoost algorithm are employed. To filter environmental noise, a combination of the least squares method and dispersion analysis is utilized to filter out irrelevant and erratic operational data. Following this, the random forest algorithm is applied to rank the significance of characteristic parameters, ensuring that only the most relevant features are selected for fault diagnosis. The BP neural network, integrated with expert knowledge, is then used to extract fault characteristics, improving model accuracy. To further refine fault detection and reflect the hydro-power unit’s real-time operation, the XGBoost algorithm is employed for fault identification. A case study demonstrates the model’s ability to predict fault characteristics 16 h in advance, confirming the effectiveness and reliability of the proposed diagnostic approach.