A General Spatiotemporal Imputation Framework for Missing Sensor Data
A General Spatiotemporal Imputation Framework for Missing Sensor Data
复制标题
DOI：
10.1109/cai54212.2023.00032
复制
发表时间：
2023-06
期刊：
影响因子：
--
通讯作者：
Aabila Tharzeen;Sai Munikoti;P. Prakash;J. Kim;Balasubramaniam Natarajan
中科院分区：
文献类型：
--
作者：
Aabila Tharzeen;Sai Munikoti;P. Prakash;J. Kim;Balasubramaniam Natarajan
精准农业、环境监测和交通网络的许多应用都依赖于在大地理区域内跨越空间和时间收集的数据。在这些时空数据库中，数据缺失可能是一个重大问题，因为它会降低下游数据分析、推理和控制算法的准确性。通过利用固有的空间关系和时间模式，数据输入或缺失数据的估计可以帮助填补这些空白。然而，现有的估计这些缺失信息的方法并不能有效地捕获时空数据结构的所有维度，从而导致错误的预测和较差的性能。在本文中，我们引入了一个通用框架，利用由传感器网络图和时间传感器数据构建的时空图来捕获联合的时空依赖关系。具体来说，我们提出了一个基于图神经网络的模型，结合递归神经网络来计算缺失信息，并证明了我们的方法对下游任务的有效性。在交通传感器网络上进行的实验表明，与最先进的imputation框架相比，该方法提高了imputation精度，平均绝对误差降低了69%，均方根误差降低了61%。
Many applications from precision agriculture, environmental monitoring and transportation networks rely on data collected across space and time over a large geographic area. Missing data can be a significant issue in these spatiotemporal databases, as it can reduce the accuracy of downstream data analysis, inferencing and control algorithms. Data imputation or the estimation of missing data can help fill these gaps by utilizing inherent spatial relationships and temporal patterns. However, existing approaches for estimating this missing information do not effectively capture all dimensions of the spatiotemporal data structure, resulting in erroneous predictions and poor performance. In this paper, we introduce a general framework that leverages a spatiotemporal graph constructed from the sensor network graph and temporal sensor data to capture the joint space-time dependencies. Specifically, we propose a graph neural network-based model in conjunction with a recurrent neural network to impute missing information and demonstrate the effectiveness of our approach for downstream tasks. Experiments on a traffic sensor network reveal enhanced imputation accuracy and up to 69% reduction in mean absolute error and 61% reduction in root mean square error compared to state-of-the-art imputation frameworks.