Abstract: New hydropower energy has been faced with new opportunities and challenges against the backdrop of the strategic goal of peaking carbon emissions and achieving carbon neutrality proposed in the national 14th Five-Year Plan. As the scale of hydropower stations expands, traditional manual inspection combined with industrial monitoring often faces more problems such as inability to automatically identify and judge faults, and low sensitivity to information feedback. This paper describes a new method for applying variational mode decomposition and image grayscale processing techniques to an analysis of the operating data of hydropower plant units, through combining 5G technology and artificial intelligence. The results show that the fractal dimension of the images features two typical frequencies of 30 Hz and 85 Hz, with the corresponding amplitudes of 0.02 and 0.009 respectively, which are detected as the dominant and secondary frequencies, far stronger than other clutter frequencies. The VMD method successfully decomposes the signals of pressure pulsation at each monitoring point so as to obtain the characteristics of various modal functions in time and frequency domains. By examining the VMD decomposition results for two monitoring points at the tail water pipe, we have found that their frequency components are similar and consistent with those monitored inside the volute. This study would provide important support for construction of intelligent hydropower stations, along with an effective means for their operation and maintenance.

Key words: hydropower stations, intelligent inspection, fault diagnosis, variational mode decomposition, image recognition