Spatial-temporal sensor data imputation in traffic data modelling

The traffic data corrupted by missing data significantly limit the robustness of traffic modelling. Our research aims to develop modelling techniques for data imputation tasks. Our first work is to impute speed data by an RBF based fitting approach for multivariate data matrixes with irregular locat...

Full description

Saved in:
Bibliographic Details
Main Author: Nie, Helei
Other Authors: Zheng Jianmin
Format: Thesis-Master by Research
Language:English
Published: Nanyang Technological University 2023
Subjects:
Online Access:https://hdl.handle.net/10356/167860
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:The traffic data corrupted by missing data significantly limit the robustness of traffic modelling. Our research aims to develop modelling techniques for data imputation tasks. Our first work is to impute speed data by an RBF based fitting approach for multivariate data matrixes with irregular location graphs. The results show that RBF approach is capable of imputing missing values, especially in a short missing span. By incorporating the available dimensions of the data, RBF method can utilize the information from time and space dimensions for effective imputation with higher missing ratios. Our second work is to propose a deep learning imputation model with a Stacking STGCN Auto-Encoder structure. We develop an adaptive graph convolution technique for to effectively utilize data missing status when traffic data suffer from severe missing conditions. The experiments show promising results when the traffic data contain large random missing regions in the time domain.