Abstract:Due to the global trend towards urbanization, people increasingly move to and live in cities that then continue to grow. Traffic forecasting plays an important role in the intelligent transportation systems of cities as well as in spatio-temporal data mining. State-of-the-art forecasting is achieved by deep-learning approaches due to their ability to contend with complex spatio-temporal dynamics. However, existing methods assume the input is fixed-topology road networks and static traffic time series. These assumptions fail to align with urbanization, where time series are collected continuously and road networks evolve over time. In such settings, deep-learning models require frequent re-initialization and re-training, imposing high computational costs. To enable much more efficient training without jeopardizing model accuracy, we propose the Topological Evolution-aware Framework (TEAM) for traffic forecasting that incorporates convolution and attention. This combination of mechanisms enables better adaptation to newly collected time series, while being able to maintain learned knowledge from old time series. TEAM features a continual learning module based on the Wasserstein metric that acts as a buffer that can identify the most stable and the most changing network nodes. Then, only data related to stable nodes is employed for re-training when consolidating a model. Further, only data of new nodes and their adjacent nodes as well as data pertaining to changing nodes are used to re-train the model. Empirical studies with two real-world traffic datasets offer evidence that TEAM is capable of much lower re-training costs than existing methods are, without jeopardizing forecasting accuracy.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in traffic prediction: 1. **Adapting to the continuously evolving road network (RN)**: Existing traffic prediction methods assume that the input road network is static, that is, the topological structure is fixed. However, in the process of urbanization, the road network will change over time, such as newly added or abandoned road nodes and edges. This evolution causes existing models to need frequent re - initialization and re - training, thereby increasing the computational cost. To solve this problem, the author proposes a framework that can adapt to the evolution of the road network. 2. **Efficiently learning complex spatio - temporal correlations on small - scale datasets**: Traditional traffic prediction models usually rely on large - scale datasets to achieve high - precision prediction, but in practical applications, the cost of constructing and maintaining these models is very high. Especially in the case of road network evolution, the amount of data for transfer learning is much less than the amount of data required for re - initializing the model. Therefore, how to effectively learn complex spatio - temporal correlations on small - scale datasets is a challenge. ### Solutions To address the above problems, the author proposes the **TEAM (Topological Evolution - aware Framework)** framework, which specifically includes the following aspects: 1. **Continual Learning Module**: - By introducing a continual learning module based on the replay mechanism, this module stores a limited number of historical data samples as a buffer. - Use the Wasserstein distance to measure the similarity between historical data and new data, and select the most stable (with the least change) node data for replay to consolidate historical knowledge and prevent forgetting. - For the data of nodes with large changes and their adjacent nodes, use the new data to update the model. - Construct a smaller new adjacency matrix, which only contains the data of the most stable nodes, the nodes with the largest changes, and the newly added nodes, thereby significantly reducing the complexity of model training. 2. **CAST model (Convolution Attention for Spatio - Temporal)**: - Propose a hybrid architecture that combines convolution and attention mechanisms for spatial and temporal calculations. - The convolution layer focuses on local patterns (such as seasonality, temporal changes, and spatially neighboring nodes), while the attention mechanism captures global patterns (such as temporal trends and far - neighboring nodes). - By combining these two mechanisms, the CAST model can more effectively learn dynamic and nonlinear spatio - temporal correlations on small - scale datasets and improve prediction accuracy. ### Main contributions - **Problem definition**: For the first time, clearly propose the problem of traffic prediction in the evolving road network. - **CAST framework**: Propose a hybrid architecture that can effectively learn on small - scale datasets. - **Continual learning module**: Enable the prediction model to effectively adapt to the evolving road network. - **Empirical study**: Through extensive experiments, verify the superior performance of the proposed framework in terms of accuracy and running time. In summary, this paper aims to solve the two major problems encountered in traffic prediction through the TEAM framework, namely, adapting to the continuously evolving road network and efficiently learning complex spatio - temporal correlations on small - scale datasets.

TEAM: Topological Evolution-aware Framework for Traffic Forecasting--Extended Version

A Deep Learning Framework About Traffic Flow Forecasting for Urban Traffic Emission Monitoring System

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

An Effective Dynamic Spatio-temporal Framework with Multi-Source Information for Traffic Prediction

Deep Temporal Convolutional Networks for Short-Term Traffic Flow Forecasting

Efficient Adaptive Spatial-Temporal Attention Network for Traffic Flow Forecasting

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

A Graph-Based Temporal Attention Framework for Multi-Sensor Traffic Flow Forecasting

Context-Aware Traffic Prediction Framework Based on Series Decomposition

Adaptive and Interactive Multi-Level Spatio-Temporal Network for Traffic Forecasting

Decoupled Dynamic Spatial-Temporal Graph Neural Network for Traffic Forecasting

Spatial-Temporal Hypergraph Neural Network for Traffic Forecasting

A Learning-Based Multimodel Integrated Framework for Dynamic Traffic Flow Forecasting

Adaptive spatial-temporal graph attention networks for traffic flow forecasting

Continual Traffic Forecasting via Mixture of Experts

Domain Adversarial Spatial-Temporal Network: A Transferable Framework for Short-term Traffic Forecasting across Cities

Topology augmented dynamic spatial-temporal network for passenger flow forecasting in urban rail transit

EGFormer: An Enhanced Transformer Model with Efficient Attention Mechanism for Traffic Flow Forecasting

A Multi-Channel Spatial-Temporal Transformer Model for Traffic Flow Forecasting

Dynamic Spatiotemporal Straight-Flow Network for Efficient Learning and Accurate Forecasting in Traffic

Spatial-Temporal Multi-Head Attention Networks for Traffic Flow Forecasting.