熊政辉, 李小军, 戴志军, 陈苏. 2019: 基于L1范数正则化的强震动加速度记录基线漂移识别方法. 地震学报, 41(1): 111-123. DOI: 10.11939/jass.20180072
引用本文: 熊政辉, 李小军, 戴志军, 陈苏. 2019: 基于L1范数正则化的强震动加速度记录基线漂移识别方法. 地震学报, 41(1): 111-123. DOI: 10.11939/jass.20180072
Xiong Zhenghui, Li Xiaojun, Dai Zhijun, Chen Su. 2019: A method for identifying the baseline drift of strong-motion records based on L1-norm regularization. Acta Seismologica Sinica, 41(1): 111-123. DOI: 10.11939/jass.20180072
Citation: Xiong Zhenghui, Li Xiaojun, Dai Zhijun, Chen Su. 2019: A method for identifying the baseline drift of strong-motion records based on L1-norm regularization. Acta Seismologica Sinica, 41(1): 111-123. DOI: 10.11939/jass.20180072

基于L1范数正则化的强震动加速度记录基线漂移识别方法

A method for identifying the baseline drift of strong-motion records based on L1-norm regularization

  • 摘要: 本文提出了一种基于L1范数正则化的基线校正新方法,即以拟合速度时程误差最小为目标,以基线漂移本身尽可能小为约束条件,经过凸优化多次迭代自动求解出满足条件的基线漂移,避免了人为选取基线漂移分段次数和基线漂移起止时刻的主观干扰;随后利用该方法对多组加入了基线漂移噪声模型的强震动加速度记录进行验证。结果表明:本文方法对于识别和处理单段式、两段式和多段式的基线漂移噪声具有普适性,能敏锐地捕捉到速度时程发生漂移的趋势(斜率变化),无需预先设定加速度基线漂移模型也可有效地识别出多种基线漂移噪声的起止位置和漂移程度;地震记录事前部分对本文方法处理结果影响较大,当记录事前部分足够长时(如20 s),识别基线漂移噪声的准确性较高,位移时程可以较好地与原始位移匹配;而对于发生漂移的速度时程,本文方法可以不受地震事前部分长短的干扰,甚至在加速度记录出现明显丢头现象时,也能很好地实现峰值速度和整个速度时程的恢复。

     

    Abstract: To identify the accurate baseline drift in ground acceleration, velocity, and displacement time series is one of the basic and challenge problems in the research of strong ground motion. This study proposes a new baseline-correction method based on L1-norm regularization. It aims at minimizing the error of fitting velocity trace subject to let the sum of absolute values of acceleration baseline drift be small. As the baseline-offset is figured out by the convexity-optimized tool automatically in this L1-norm regularization based baseline-correction method, the subjective interferences can be well avoided such as selecting segmentation times and the start and end moments. And then representative noise models of acceleration baseline offset are added respectively to typical strong-motion records in order to test and verify the new method. The results shows that our method is universal for identifying and processing single-, double-, and multi-stage baseline drift noises. It can sensitively capture the trend (slope) change of the velocity trace while it’s no need to set segmentation times and positions of piecewise linear fitting in advance. The pre-event interval of strong-motion record has a great influence on the processing results of this method. If the pre-event interval is long enough (e.g. 20 seconds) in a record, the identification of the baseline drift noise will be much more accurate, and the recovered displacement trace will match better with the real one. Additionally, this method shows good performance to recover peak ground velocity and the whole velocity time series even if the record almost has no pre-event portion.

     

/

返回文章
返回