基于L-M算法的神经网络在环境振动分析中消除本底振动的应用

Application of a Neural Network based on L-M Algorithm in environmental vibration analysis to remove background vibration

公路、铁路及城市轨道交通引起的环境振动实测数据中含有本底振动的干扰,从时域分析角度提出基于L-M(Levenberg-Marquardt)算法的神经网络法消除本底振动,阐述了该法的基本原理和实现步骤,采用L-M算法对神经网络进行训练,具有收敛速度快、计算精度高的特点。通过一段交通振动加速度时程与一段本底振动加速度时程叠加合成实测振动加速度时程,分别用L-M神经网络法和其他几种方法对合成的实测振动加速度时程进行本底振动消除计算和对比分析。计算结果表明,L-M神经网络法能更加精确的计算出真实交通振动产生的时程曲线、功率谱密度曲线、1/3倍频程中心频率处振动加速度级和计权振级。

Background vibration can disturb environmental vibration induced by highway,railway and urban rail transportation. The neural network method based on L-M（ Levenberg-Marquardt） algorithm,a time-domain analysis approach,was proposed to remove background vibration in environmental vibration testing data. The basic principle and implementation steps were presented. A neural network was trained using L-M algorithm to speed up the network training rate and improve the accuracy of network training. A background vibration acceleration time history was superimposed on a transportation vibration acceleration time history to synthesize a tested vibration acceleration time history. It was used to remove background vibration with L-M neural network approach and other approaches. The calculated results indicated that L-M neural network method can be used to calculate time history,power spectral density,vibration acceleration level on the one-third octave band center frequency and weighted level of a true transportation vibration more accurately. It was shown that L-M Neural network method is superior to other current methods.