交通流数据时空相关修复方法评价分析

Evaluating Alternative Temporal-spatial Correlated Imputation Methods for Traffic Flow Data

文中选取了预测均值匹配、K近邻和贝叶斯时间矩阵分解三种时空相关数据修复方法,基于上海市中环路交通流数据,以时间相关方法为基准,评价分析不同缺失模式下不同方法的修复效果.结果表明:K近邻方法可有效追踪交通流的变化,修复精度和稳定性较好,时间相关和贝叶斯时间矩阵分解方法在特定场景可能失效.引入空间信息可有效提升修复精度,相邻检测器、上游检测器和下游检测器的数据对修复效果提升的贡献依次递减.

Based on the selected three methods of time-space correlation data restoration:prediction mean matching,K nearest neighbor and Bayesian time matrix decomposition,according to the traffic flow data of Shanghai Zhonghuan Road,and taking the time correlation method as the benchmark,the restoration effects of different methods under different missing modes were evaluated and analyzed.The results show that the K-nearest neighbor method can effectively track the change of traffic flow,and the imputation accuracy and stability are good,and the time correlation and Bayesian time matrix decomposition methods may fail in certain scenes.The introduction of spatial information can effectively improve the restoration accuracy,and the contribution of the data of adjacent detectors,upstream detectors and downstream detectors to the improvement of the restoration effect decreases in turn.