Fault Diagnosis Based on HBA-MCKD-MFFR for Hydroelectric Power Units

The Honey Badger Algorithm-Maximum Correlation Kurtosis Deconvolution with Multi-scale Feature Fusion ResNet (HBA-MCKD-MFFR) is proposed in this research as the solution to the problem of improper feature of failure signals in the original hydropower unit due to vibration of the external environment and distortion. First, this research come up with an HBA-MCKD improvement approach where HBA algorithm is utilized to find the optimal parameters of Maximum Correlation Kurtosis Deconvolution (MCKD). Subsequently, a diagnosis model with attentive multi-scale feature fusion ResNet is presented to account for local feature contributions. Multiscale residual blocks can be utilized where the features retrieved at different scales feed into MFFR for training and classification. Experiment results will demonstrate that this method performs well, with the fault type classification accuracy of 99.23% exceeding other top algorithms.