A Novel Method for Air Quality Data Imputation by Nuclear Norm Minimization

摘要: Missing data is a frequently encountered problem in environment research community. To facilitate the analysis and management of air quality data, for example, PM2.5 concentration this study, commonly adopted strategy handling missing values samples to generate complete set using imputation methods. Many methods based on temporal or spatial correlation have been developed purpose existing literatures. The difference various lies characterizing dependence relationship with different mathematical models, which crucial imputation. In paper, we propose two novel principled nuclear norm matrix since it measures such global fashion. first method, termed as minimization (GNNM), tries impute through directly minimizing whole sample matrix, thus at same time maximizing linear samples. second called local (LNNM), concentrates more each its most similar are estimated from results method. way, can be performed those highly correlated instead GNNM, reducing adverse impact irrelevant evaluated measured every 1 h by 22 monitoring stations. simulated percentages. imputed compared ground truth evaluate performance experimental verify effectiveness our methods, especially LNNM,