基于Lasso回归及模型修正的双重回归缺失值插补方法研究

Research on Missing Value Interpolation Method of Double Regression Based on Lasso Regression and Model Modification

针对各传感网络中传感数据因工作环境变化、传感设备异常等因素而引起的测量值缺失的问题,提出了一种基于Lasso回归及模型修正的双重回归缺失值插补方法。该方法采用原始数据滑动窗口法生成数据集并随机删除部分数据,以Lasso回归模型为基准,使用岭回归与皮尔逊相关性分析联合分析且生成集成岭回归与相关性的数据集,并将其作为Lasso回归模型的特征(双列),以双重回归方式进行模型修正,最终实现对缺失值的插补。以西储大学轴承数据为例,对所提方法及另外2种缺失值插补方法(KNN的数据插补和Lasso回归的缺失值插补)在缺失率为4%、10%和20%下进行比较,并采用均方根误差、模型训练时间及决定系数作为评估指标。结果表明,基于Lasso回归及模型修正的双重回归缺失值插补方法具有较好的表现,为后续的故障诊断提供可靠的基础数据。

Aiming at the missing measurement values caused by the change of working environment and abnormal sensor equipment of petrochemical units,a method of double regression of missing value interpolation based on Lasso regression and model modification is proposed.The method uses the original data sliding window method to generate the dataset and randomly delete some data.Taking Lasso regression model as a benchmark,the method uses ridge regression with Pearson correlation analysis to jointly analyze and generate a dataset integrating ridge regression and correlation.This is a feature(double row)of the Lasso regression model.On the basis of the dual regression,the missing value is finally interpolated.Taking the bearing data from Western Reserve University as an example,the proposed method and the other two missing value interpolation methods(KNN data interpolation,Lasso regression missing value interpolation)are compared at missing rate of 4%,10%and 20%,and the root mean square error,model training time and determination coefficient are used as evaluation indexes.The results show that the double regression missing value interpolation method based on Lasso regression and model modification performs well,and provides reliable basic data for subsequent fault diagnosis.