论文部分内容阅读
针对高轨航天器非平稳随机振动信号模态频率密集,传统时变信号处理方法难以准确计算时变功率谱密度,从而影响地面对航天器操作决策的特点,提出了基于经验模式分解(EMD)的时变自回归(TVAR)多分量过程神经元网络(PNN)模型。该方法利用过程神经网络自适应跟踪一个时变系统,从而获得具有时变系数的参数化数学模型,进而可得到信号的时变参数谱。首先通过 EMD 对原始时间序列进行分解, 使之成为不同尺度的本征模函数(IMF), 然后利用基函数展开的 TVAR 过程 PNN 对每个 IMF 分别进行时变参数分析并以此确定其时变自功率谱密度,最后所有分量的时变自功率谱密度通过线性叠加进行重构, 作为原始信号的时变自功率谱密度。仿真和实例计算结果表明:这种方法不仅简单、有效和可行,而且频率分辨性能优于传统的分析方法。
Aiming at the characteristics of high space orbit random vibration signal of high-orbit spacecraft, and the traditional time-varying signal processing method, it is difficult to calculate the time-varying power spectral density accurately and thus affect the operation decision of the spacecraft. Based on the empirical mode decomposition (EMD) ) Time-varying autoregressive (TVAR) multi-component process neural network (PNN) model. This method uses process neural network to adaptively track a time-varying system, so as to obtain a parametric mathematical model with time-varying coefficients, and then the time varying parameter spectrum of the signal can be obtained. Firstly, EMD decomposes the original time series into IMFs of different scales, and then uses the TVAR process PNN developed by the basis function to analyze the time-varying parameters of each IMF separately and determine the time-varying Self-power spectral density, the last of all components of time-varying self-power spectral density by linear superposition reconstruction, as the original signal time-varying self-power spectral density. The simulation and example calculation results show that this method is not only simple, effective and feasible, but also the frequency resolution performance is better than the traditional analysis methods.