Adaptive filtering method and application based on RNN
REN Hongyi1, LIU Xiangyu1, XIAN Ganling1, LAN Jingyan1,2
1. College of Civil Engineering and Architecture, Guilin University of Technology, Guilin 541004, China;
2. Guangxi Provincial Key Lab of Geomechanics and Geotechnical Engineering, Guilin University of Technology, Guilin 541004, China
摘要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，本文结合递归最小二乘算法（RLS）和和循环神经网络（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，运用本文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。
Abstract：At present, there are some defects in seismic wave filtering methods of earthquake engineering research field, such as human experience interference, peak spike, noise interference, etc. In this paper, a new adaptive filtering method is proposed by combining recursive least squares (RLS) algorithm and recurrent neural network (RNN) model. The results show that the new method performs filtering by setting adaptive adjustment filter parameters and the self-iteration algorithm. It is superior to the traditional filtering method recommended by the United States Geological Survey (USGS) in noise recognition ability and filtering speed, and can effectively reduce the distortion, damage and phase advance of the original waveform after filtering. At the same time, the adaptive filtering method was applied to near-field seismic records containing velocity pulses at different site classification of stations. The adaptive filtering method has been further proven to be effective. The research results provide a new idea and method for filtering analysis in the field of Earthquake engineering, and can also provide reference for seismic record processing and related applications.
任鸿燚1，刘翔宇1，咸甘玲1，兰景岩1,2. 基于循环神经网络（RNN）的自适应滤波方法及应用研究[J]. 振动与冲击, 2024, 43(7): 327-333.
REN Hongyi1, LIU Xiangyu1, XIAN Ganling1, LAN Jingyan1,2. Adaptive filtering method and application based on RNN. JOURNAL OF VIBRATION AND SHOCK, 2024, 43(7): 327-333.
[1] 何峰, 周亚同, 赵翔宇, 石超君. 基于波原子变换的三维地震信号盲去噪[J]. 振动与冲击, 2019, 38(8):88-95.
He Feng, Zhou Yatong,, Zhao Xiangyu, Shi Chaojun. Blind denoising of 3D seismic signals based on the wave atom transform [J]. Journal of Vibration and Shock, 2019, 38(8): 88-95.
[2] 熊永学,晁元萍,康平等.柴达木盆地小组合基距地震资料干扰波分析及去噪对策[J].石油地球物理勘探, 2002(S1): 66-70+221.
Xiong Yongxue, Chao Yuanping, Kang Ping et al. Interference wave analysis and denoising strategy for seismic data of small base interval in Qaidam Basin [J]. Oil Geophysical Prospecting,2002(S1):66-70+221.
[3] TRIFUNUC M D. Zero baseline correction of strong-motion accelerograms [J]. Bulletin of the Seismological Society of America, 1971, 61(5): 1201-1211.
[4] CONVERSE A M, BRADY A G. BAP: Basic strong-motion accelerogram processing software; version 1.0[J]. Center for Integrated Data Analytics Wisconsin Science Center, 1992.,42(237):37-41
[5] BOORE D M, BOMMER J J. Processing of strong-motion accelerograms: needs, options and consequences [J]. Soil Dynamics and Earthquake Engineering, 2005, 25(2): 93-115.
[6] GRAIZER V M. Determination of the true ground displacement by using strong motion records [J]. Earth Physics, 1979, 15(2):875-885.
[7] Yang J, Li J B, Lin G. A simple approach to integration of acceleration data for dynamic soil–structure interaction analysis [J]. Soil Dynamics and Earthquake Engineering, 2006, 26(8): 725-734.
[8] Xu Xin, He Hangen, Hu Dewen. Efficient reinforcement learning using recursive least-squares methods [J]. Journal of Artificial Intelligence Research, 2002, 16: 259-292.
[9] 王涛,翟绪恒,孟丽岩.在线自适应神经网络算法及参数鲁棒性分析[J].振动与冲击, 2019, 38(8): 210-217.
Wang Tao, Zhai Xuheng, Meng Liyan. Online Adaptive Neural Network Algorithm and Parameter Robustness Analysis [J]. Vibration and Shock, 2019, 38(8):210-217.
[10] 杨丽,吴雨茜,王俊丽,刘义理.循环神经网络研究综述[J].计算机应用,2018,38(S2):1-6+26.
Yang Li, Wu Yuqian, Wang Junli, Liu Yili. Review of research on cyclic Neural Networks [J]. Computer Applications, 2018, 38(S2): 1-6+26.
[11] 李超,柴玉梅,南晓斐,等.基于深度学习的问题分类方法 研究[J].计算机科学,2016,43(12):115-119．
Li Chao, Chai Yumei, Nan Xiaofei, et al. Research on Problem Classification Based on Deep Learning [J]. Computer Science,2016,43(12):115-119.
[12] Dongwei Chen, Fei Hu, Guokui Nian, Tiantian Yang. Deep Residual Learning for Nonlinear Regression [J]. Entropy, 2020, 22: 193.
[13] 庞荣. 深度神经网络算法研究及应用[D]. 成都：西南交通大学, 2016.
Pang Rong. Research and Application of Deep NeuralNetwork Algorithm [D]. Chengdu: Southwest Jiaotong University, 2016.
[14] 张静,农昌瑞,杨智勇.基于卷积神经网络的目标检测算法综述[J].兵器装备工程学报, 2022,43(6):37-47.
Zhang Jing, Nong Chang-rui, Yang Zhi-yong. Review of object Detection Algorithm Based on Convolutional Neural Network [J]. Journal of Ordnance Equipment Engineering, 2022, 43(6): 37-47.
[15] Chessel D. The spatial autocorrelation matrix [M]. Vegetation dynamics in grasslans, healthlands and mediterranean ligneous formations. Springer, Dordrecht, 1981: 177-180.
[16] Tylavsky D J, Sohie G R L. Generalization of the matrix inversion lemma [J]. Proceedings of the IEEE, 1986, 74(7): 1050-1052.
[17] Pascanu R, Mikolov T, Bengio Y. On the difficulty of training recurrent neural networks [C]. Proceedings of the 30th International conference on Machine Learning. Atlanta, Georgia, USA 2013.
[18] Jaeger H. A tutorial on training recurrent neural networks, covering BPPT, RTRL, EKF and the" echo state network" approach [R]. GMD Report 159, German National Research Center for Information Technology, 2002.
[19] 刘铁林, 孙宇城, 张柔佳. PEER地震动数据库中含速度脉冲近场地震记录[J].防灾减灾工程学报, 2018, 38(2): 367-372.
Liu Tie-Lin, Sun Yu-Cheng, Zhang Rou-Jia. Near-field seismic records with velocity pulse in PEER ground motion database [J]. Journal of Disaster Prevention and Mitigation Engineering, 2018, 38(2):367-372.
[20] NEHRP recommended provisions for seismic regulations for new buildings and Other Structures[S]. FEMA 450, Part 1 (Provisions) and Part 2 (Commentary), Developed for the Federal Emergency Management Agency, Washington, DC, Building Seismic Safety Council (BSSC). 2004.
