基于机器学习的随机信号缺失数据功率谱估计

Power Spectrum Estimation of Random Signal Missing Data Based on Machine Learning

为了解决随机信号在传递接收过程中可能出现的数据不规则、不完整的问题,提出了一种极值梯度提升树-支持向量回归(XGBoost-SVR)与门控循环单元神经网络结合的预测方法,用于稀疏随机信号的缺失数据填补。先使用极值梯度提升树和支持向量回归结合对原始数据集填补,再使用门控循环单元模型进行训练并预测出缺失的数据,最后通过傅里叶变换和小波变换,实现功率谱估计。数值分析结果表明:在含有噪声干扰的情况下,即使数据缺失高达70%,该预测方法仍然可以很好地还原目标功率谱。该方法处理的缺失数据可以更好地还原随机信号的特征,为信号预测和灾害预防提供有力支持。

In order to solve the problem of irregular and incomplete data that may occur during the transmission and reception of random signals,a prediction method based on the combination of XGBoost-SVR and gated recurrent unit neural network was proposed for use missing data filling for sparse random signals.Firstly,the original dataset was filled using an extreme gradient boosting tree and support vector regression.Further the gated cyclic unit model was used to train and predict the missing data.Finally,the power spectrum estimation was achieved by Fourier transform or wavelet transform.The numerical analysis results showed that in the presence of noise interference,even if the data is missing by up to 70%,the proposed method can reproduce the target power spectrum well.The missing data processed by this method can better restore the characteristics of random signals,providing strong support for signal prediction and disaster prevention.