An Algorithm Of Imputing Missing Network Data Based On Matrix Low-rank Decomposition

Reliable and complete traffic data is the basis and premise of Intelligent Transportation Systems,such as traffic condition identification and traffic induction,Detectors in the traffic network are broken often because of the environment and working for a long time,which leads to the loss of traffic flow data.To solve the problem of structural missing of traffic flow data,it is meaningful how to make use of the existing theory system to build a suitable data imputation model and recover the loss data effectively.Aiming at the problem that Traditional matrix imputation model cannot repair the structural missing traffic flow data accurately,a data imputation model called SVR-LRMD-TDMC is proposed combined LRMD-TDMC(a Traffic Data Matrix Complete algorithm based on Low-Rank Matrix Decomposition)with SVR(Support Vector Regression).The preprocessed traffic flow data matrix named training matrix is decomposed into a low-rank matrix and a sparse matrix.The training matrix is correlated to the low-rank matrix with a coefficient matrix by the Low-Rank Representation(LRR).The coefficient matrix can be solved with the ALM algorithm.The missing data can be recovered by the coefficient matrix and a new dictionary matrix,which is the spatial repairing result.The final result is a weighted sum of the spatial repairing result and the temporal predicted by SVR.The proposed model is tested with the real traffic data collected from Furong district in Changsha.The result shows that the average reconstruction error of the LRMD-TDMC model decreases 3.73 percentage point compared with ARIMA,PPCA and SVR.The SVR-LRMD-TDMC model decreases 1.95 percentage point compared with the LRMD-TDMC model.A conclusion is drawn that the SVR-LRMD-TDMC and LRMD-TDMC are efficient algorithm on imputing the structural missing traffic data.