For enhancing both accuracy and real-time capability of damage identification for UHV converter station equipment in seismic period

in this paper a damage identification method of equipment based on long short-term memory (LSTM) networks combined with wavelet scattering feature extraction is proposed. The real data containing different damage conditions is formed by simulating the acceleration response of the converter station equipment in the seismic period through finite element simulation. The wavelet scattering technology is used was to extract features from the acceleration signals so to effectively reduce noise and preserve damage-related features. These extracted features are then input into the LSTM model for damage identification. The results show that the LSTM network based on the wavelet scattering features significantly improves both speed and accuracy of damage identification compared to that directly using raw acceleration data

and the final identification accuracy of the model is up to 95%. This method improves effectively both accuracy and efficiency of seismic damage identification of converter station equipment

provides reliable technical support for health monitoring and post-disaster assessment of such power infrastructure as converter station and has broad engineering application potential.