Abstract:Traffic flow forecasting is a fundamental research issue for transportation planning and management, which serves as a canonical and typical example of spatial-temporal predictions. In recent years, Graph Neural Networks (GNNs) and Recurrent Neural Networks (RNNs) have achieved great success in capturing spatial-temporal correlations for traffic flow forecasting. Yet, two non-ignorable issues haven't been well solved: 1) The message passing in GNNs is immediate, while in reality the spatial message interactions among neighboring nodes can be delayed. The change of traffic flow at one node will take several minutes, i.e., time delay, to influence its connected neighbors. 2) Traffic conditions undergo continuous changes. The prediction frequency for traffic flow forecasting may vary based on specific scenario requirements. Most existing discretized models require retraining for each prediction horizon, restricting their applicability. To tackle the above issues, we propose a neural Spatial-Temporal Delay Differential Equation model, namely STDDE. It includes both delay effects and continuity into a unified delay differential equation framework, which explicitly models the time delay in spatial information propagation. Furthermore, theoretical proofs are provided to show its stability. Then we design a learnable traffic-graph time-delay estimator, which utilizes the continuity of the hidden states to achieve the gradient backward process. Finally, we propose a continuous output module, allowing us to accurately predict traffic flow at various frequencies, which provides more flexibility and adaptability to different scenarios. Extensive experiments show the superiority of the proposed STDDE along with competitive computational efficiency.

Transient loss guarantees for long-range dependent multiple time-scale traffic

Short-term Traffic Prediction with Deep Neural Networks and Adaptive Transfer Learning

Bandwidth Management for Guaranteeing Transient Loss Performance of Real-Time Communications

Long-range-dependent Analysis Methods of Self-similar Network Traffic

On the LRD of the Aggregated Traffic Flows in High-Speed Computer Networks

Scaling behavior of Internet packet delay dynamics based on small-interval measurements

Impact of Short-range Dependence on the Performance of Network Analysis Methods

The Transient Analysis of Loss Networks

Transfer Learning Based High-Speed Link Transient Modeling Method

Large-Scale Deterministic Networks: Architecture, Enabling Technologies, Case Study and Future Directions

Two-Timescale Resource Management for Ultrareliable and Low-Latency Vehicular Communications

A Truncated alpha-Stable Long-Range Dependent Process Model for Modeling Network Traffic

Transient loss performance of a class of finite buffer queueing systems

Temporal characteristics and reliability analysis of railway transportation networks

Time‐varying Network Link Loss Rate Tracking from Unicast End‐to‐end Measurements

Traffic Dynamics Considering Packet Loss in Finite Buffer Networks

Unveiling Delay Effects in Traffic Forecasting: A Perspective from Spatial-Temporal Delay Differential Equations

Temporal Dependence Network Link Loss Inference from Unicast End-to-End Measurements

Measurement, Modeling, and Analysis of TCP in High-Speed Mobility Scenarios

An Efficient Modeling and Dimensioning Approach for Integrated Streaming and Elastic Traffic

Reliable Multi-Path Routing with Bandwidth and Delay Constraints