Health condition monitoring of offshore substation based on time-frequency analysis and deep cluster-based autoencoder

A health monitoring method for offshore substation based on time-frequency denoising feature extraction and deep clustering autoencoders is proposed to address the issue of structural damage in offshore substations caused by harsh marine environments and complex wind loads. This method utilizes monitoring data such as tilt angles, vibration accelerations and jacket strains. Vibration signals are denoised using a variational mode decomposition method that accounts for multi-channel coherence, and time-frequency images of the denoised signals are extracted through continuous wavelet transform. Subsequently, a deep convolutional clustering autoencoder is employed to learn the latent features of the monitoring data fro m offshore substations, characterizing the health states of substations under different operating conditions at the feature level. Finally, an anomaly detection strategy is constructed by combining clustering confidence indices and reconstruction error metrics, enabling real-time health monitoring of offshore substations. The research demonstrates that this method effectively identifies subtle damages that are difficult to capture using traditional methods, improves monitoring accuracy and early warning efficiency, and provides technical support to ensure the safe operation of offshore wind farms.