Abstract:Long-term traffic flow forecasting plays a crucial role in intelligent transportation as it allows traffic managers to adjust their decisions in advance. However, the problem is challenging due to spatio-temporal correlations and complex dynamic patterns in continuous-time stream data. Neural Differential Equations (NDEs) are among the state-of-the-art methods for learning continuous-time traffic dynamics. However, the traditional NDE models face issues in long-term traffic forecasting due to failures in capturing delayed traffic patterns, dynamic edge (location-to-location correlation) patterns, and abrupt trend patterns. To fill this gap, we propose a new NDE architecture called Multi-View Neural Differential Equations. Our model captures current states, delayed states, and trends in different state variables (views) by learning latent multiple representations within Neural Differential Equations. Extensive experiments conducted on several real-world traffic datasets demonstrate that our proposed method outperforms the state-of-the-art and achieves superior prediction accuracy for long-term forecasting and robustness with noisy or missing inputs.

What problem does this paper attempt to address?

This paper attempts to solve the challenging problems in long - term traffic flow prediction, especially the difficulties brought by spatio - temporal correlations and complex dynamic patterns in continuous - time flow data. Specifically: 1. **Delayed Propagation**: Traditional models are unable to capture the delayed propagation effect of traffic conditions. For example, on a highway, a slow - moving vehicle will cause the following vehicles to slow down. This influence is delayed and the time and location are uncertain. 2. **Dynamic Spatial Dependence**: The spatial dependence relationships between different locations in the traffic network change over time, especially during peak hours or when an accident occurs. How to clearly and quantitatively capture these relationships is a difficult problem. 3. **Sudden Trend Patterns**: Local traffic flow mutations caused by unexpected events (such as car accidents or bad weather) deviate from the overall traffic situation, and existing models do not fully consider these factors. To solve these problems, the author proposes a new neural differential equation architecture, called Multi - View Neural Differential Equations (MNDE). MNDE captures different state variables (views) of the current state, the delayed state and the trend by learning multiple latent representations, so as to better handle the complex dynamic patterns in long - term traffic flow prediction. ### Main Contributions - **Multi - view Design**: The MNDE framework contains multiple NDE modules, which are respectively used to capture complex delayed propagation, dynamic spatial dependence and sudden trend patterns. - **Delayed Neural Differential Equation Module (DNDE)**: A new DNDE module is introduced to deal with the delayed propagation effect in traffic flow dynamics. - **Differentiation Module (DM)**: The sudden change patterns in traffic flow data are captured through the self - attention mechanism. - **Experimental Verification**: Extensive experiments on multiple real - world traffic data sets show that MNDE is superior to existing methods in long - term prediction accuracy and robustness to noisy or missing inputs. ### Summary The main objective of the paper is to improve the accuracy and robustness of long - term traffic flow prediction by proposing the MNDE framework, with special attention to capturing complex dynamic characteristics such as delayed propagation, dynamic spatial dependence and sudden trend patterns.

Multi-View Neural Differential Equations for Continuous-Time Stream Data in Long-Term Traffic Forecasting

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

Continuously Evolving Graph Neural Controlled Differential Equations for Traffic Forecasting

Short-Term Forecasting of Traffic Flow and Speed: A Deep Learning Approach

Adaptive Decision Spatio-temporal neural ODE for traffic flow forecasting with Multi-Kernel Temporal Dynamic Dilation Convolution

Graph Neural Rough Differential Equations for Traffic Forecasting

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

Temporal Super-Resolution Traffic Flow Forecasting Via Continuous-Time Network Dynamics

TrafficStream: A Streaming Traffic Flow Forecasting Framework Based on Graph Neural Networks and Continual Learning

Spatial-Temporal Hypergraph Neural Network for Traffic Forecasting

Graph-based Multi-ODE Neural Networks for Spatio-Temporal Traffic Forecasting

Decoupled Dynamic Spatial-Temporal Graph Neural Network for Traffic Forecasting

GDENet: Graph Differential Equation Network for Traffic Flow Prediction

Link prediction by continuous spatiotemporal representation via neural differential equations

Network-Wide Traffic Flow Dynamics Prediction Leveraging Macroscopic Traffic Flow Model and Deep Neural Networks

Deep Multi-View Channel-Wise Spatio-Temporal Network for Traffic Flow Prediction

Spatial-Temporal Graph ODE Networks for Traffic Flow Forecasting

CDGNet: A Cross-Time Dynamic Graph-based Deep Learning Model for Traffic Forecasting

Forecast Network-Wide Traffic States for Multiple Steps Ahead: A Deep Learning Approach Considering Dynamic Non-Local Spatial Correlation and Non-Stationary Temporal Dependency

Multi-Scale Spatio-Temporal Attention Networks for Network-Scale Traffic Learning and Forecasting

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