Improvement to the sources selection to identify the low frequency noise induced by flood discharge

Recent studies indicate that low frequency noise (LFN) is generated by flood discharge of high dam. For prototype observation of the LFN, the observed data usually contain multiple sources induced by different mechanisms. To separate and identify the LFN generated by flood discharge from the multiple sources, an improved version of single-channel blind source separation (SCBSS) is proposed. In this study, the source number estimation is improved by a singular entropy (SE) method based on the eigenvalues calculated by principal components analysis (PCA). Then an SCBSS algorithm with no interruption is proposed. Both traditional method and PCA-SE method may result in extra sources that do not exist actually and are introduced by SCBSS due to the misconduct of human judgment or underestimation of threshold. Therefore, a cross-correlation procedure is proposed to identify and eliminate the extra sources and other sources that we are not really concerned about. The proposed method is first applied to a pre-determined signal to validate its effectiveness. Then the LFN data observed during the flood discharge of the Jin'anqiao hydropower station are analyzed and separated using this improved method. Two components, with dominant frequencies about 0.7 Hz and 0.95 Hz respectively, are successfully recognized as the actual acoustic sources induced by the flood discharge.