Abstract:Traffic forecasting is one of the most well-studied problems in the Intelligent Transportation Systems (ITS). However, existing studies mainly utilize Euclidean distance or road network distance as the metrics at the spatial level, where the distance between two points is calculated based on the GPS coordinates. When using the spatial level metrics to construct the road network's topological and indexing structure, two completed unrelated road segments such as the upper and lower levels of the viaduct will be contiguous in the adjacent matrix and consequently be indexed in the same block. However, it may taking a long time to drive through these two road segments, which makes the spatial proximity meaningless, and this is pretty common during the morning and evening rush hours. Thus, it is very important to take into consideration of the temporal metric along with the spatial distance metric when dealing with the traffic forecasting problem. Motivated by the above observation, we propose a novel Temporal Metrics Based Aggregated Graph Convolution Network (TMAGCN), which utilize the time reachability based road network weight metric, and reveal its unique characteristics. TMAGCN mainly consists of three components: spatio-temporal reachability calculation, region clustering(road segments aggregating), and traffic flow forecasting. We explore the cluster of road segments and the reconstruction of the spatio-temporal reachability graph with temporal-level information rather than spatial-level information for better capturing the latent region information. To the best of our knowledge, this paper is the first attempt on this problem. The comprehensive experiments indicate that TMAGCN achieves higher accuracy and efficiency than other off-the-shelf models.

Metric-Based Multi-Task Grouping Neural Network for Traffic Flow Forecasting.

Traffic Flow and Speed Forecasting Through a Bayesian Deep Multi-Linear Relationship Network.

Network Traffic Classification Via Non-Convex Multi-Task Feature Learning

Learning Common Metrics for Homogenous Tasks in Traffic Flow Prediction

Spatial-Temporal Graph Multi-Gate Mixture-of-Expert Model for Traffic Prediction

Short-Term Traffic Flow Forecasting: Multi-Metric Knn with Related Station Discovery

Daily Long-Term Traffic Flow Forecasting Based on a Deep Neural Network

Hybrid Multi-Metric K-Nearest Neighbor Regression for Traffic Flow Prediction

A Short-term Traffic Flow Forecasting Method Based on Combination of Fuzzy C-mean Clustering and Neural Network

Knowledge-Based Multiple Relations Modeling for Traffic Forecasting

Multi-task Learning for Sparse Traffic Forecasting

Deep Architecture for Traffic Flow Prediction: Deep Belief Networks with Multitask Learning

MAF-GNN: Multi-adaptive Spatiotemporal-flow Graph Neural Network for Traffic Speed Forecasting

Temporal Metrics Based Aggregated Graph Convolution Network for traffic forecasting

Multitask Hypergraph Convolutional Networks: A Heterogeneous Traffic Prediction Framework

Temporal Multi-Graph Convolutional Network for Traffic Flow Prediction

A Hybrid Method for Traffic Flow Forecasting Using Multimodal Deep Learning

A traffic flow-forecasting model based on multi-head spatio–temporal attention and adaptive graph convolutional networks

Adaptive spatial-temporal graph attention networks for traffic flow forecasting

Multi-scale feature enhanced spatio-temporal learning for traffic flow forecasting