基于循环神经网络的自适应滤波方法及应用研究

Adaptive filtering method and application based on RNN

针对目前地震工程研究领域在滤波方法上存在人为因素、峰值突刺、噪声干扰等方面的缺陷,结合递归最小二乘法(RLS)和循环神经网络(RNN)模型,提出了一种自适应滤波的新方法。研究分析表明,该方法通过设置自适应调节滤波器参数以及算法的自我迭代等方式进行滤波,对噪声识别能力和滤波速度上均优于美国地质调查局(United States Geological Survey,USGS)所推荐的传统滤波方法,并可有效降低滤波后对原始波形的失真损坏以及相位提前等问题。同时,运用所提自适应滤波方法将其应用于不同场地类型台站的含速度脉冲近场地震记录,进一步验证了自适应滤波方法的有效性和适用性。研究成果为地震工程领域的滤波分析提出了一种新思路和新方法,也可为地震记录处理及相关应用工作提供参考。

Aiming at shortcomings of filtering methods in current study fields of earthquake engineering of human factors,peak spikes and noise interference,a new adaptive filtering method combining recursive least squares(RLS)algorithm and recurrent neural network(RNN)models was proposed.Study analysis showed that by adaptively adjusting filter parameters and setting self-iteration of algorithm,the proposed method achieves better noise recognition ability and filtering speed than traditional filtering methods recommended by United States Geological Survey(USGS)do;it can effectively reduce distortion,damage and phase advance of the original waveform after filtering;meanwhile,when the adaptive filtering method proposed here is applied in near-field seismic records containing velocity pulses of seismic stations at different sites,its effectiveness and applicability can be verified;the study results can provide new idea and method for filtering analysis in seismic engineering field,and also provide a reference for seismic record processing and related applications.