Abstract:The ever increasing amount of GPS-equipped vehicles provides in real-time valuable traffic information for the roads traversed by the moving vehicles. In this way, a set of sparse and time evolving traffic reports is generated for each road. These time series are a valuable asset in order to forecast the future traffic condition. In this paper we present a deep learning framework that encodes the sparse recent traffic information and forecasts the future traffic condition. Our framework consists of a recurrent part and a decoder. The recurrent part employs an attention mechanism that encodes the traffic reports that are available at a particular time window. The decoder is responsible to forecast the future traffic condition.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to process sparse traffic data for traffic prediction. Specifically, with the increasing number of vehicles equipped with GPS, these vehicles provide valuable real - time traffic information during their travels. However, these traffic reports are sparse and change over time. Therefore, how to effectively encode these sparse traffic information and accurately predict future traffic conditions is a challenge. ### Problem Definition Given the available travel times \(T_{t - L + 1:t}\) in the past \(L\) time windows, the traffic prediction model needs to predict the travel times \(T_{t + 1:t + H}\) in the next \(H\) time windows, where the vector \(T_t\) contains the travel times of all road segments \(E\) at time \(t\). There are missing values in the input matrix \(T_{t - L + 1:t}\in\mathbb{R}^{|E|\times L}\) (that is, some road segments are not passed by any vehicle within a specific time window). The prediction matrix \(T_{t + 1:t + H}\in\mathbb{R}^{|E|\times H}\) contains the predicted travel times of all road segments \(E\) in the next \(H\) time windows. ### Solution Overview To solve this problem, the author proposes a new deep - learning framework, which consists of two parts: 1. **Recurrent Part**: Use the Attention Mechanism to encode the available traffic reports within a specific time window. 2. **Decoder**: Responsible for predicting the traffic conditions in the next time window. ### Main Contributions - **Handling Sparse Data**: Through the Attention Mechanism and Road Segment Embeddings, this framework can effectively handle sparse traffic data. - **Periodicity Modeling**: By analyzing historical travel - time reports, calculate the average travel time of each road segment in different time periods, and subtract the historical average travel time of this road segment from each travel time, thereby forcing the model to learn the deviation of each road segment in different time periods. - **Normalizing Travel Time**: In order to consider the differences in the lengths and speed limits of different road segments, the author normalizes the travel time. - **Multi - Head Attention**: Expands the traditional "Scaled Dot - Product Attention" mechanism and can better capture the correlations between different road segments. ### Application Scenarios This framework can be applied to real - time traffic prediction systems to help users choose the optimal route and improve travel efficiency. For example, given a query path \(P_q\) and departure time \(t_q\), this model can estimate the travel time required for the entire path, thereby providing users with accurate arrival - time predictions. In summary, the main objective of this paper is to develop an effective deep - learning framework that can accurately predict future traffic conditions based on processing sparse traffic data, thereby improving the service quality of traffic prediction and travel - time estimation.

A Novel Framework for Handling Sparse Data in Traffic Forecast

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

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

Geospatial Data to Images: A Deep-Learning Framework for Traffic Forecasting

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

Traffic Graph Convolutional Recurrent Neural Network: A Deep Learning Framework for Network-Scale Traffic Learning and Forecasting

Decoupled Dynamic Spatial-Temporal Graph Neural Network for Traffic Forecasting

Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting

Traffic Flow Prediction via Spatial Temporal Graph Neural Network

A Study of Deep Learning for Network Traffic Data Forecasting

A Deep Learning Framework for Traffic Data Imputation Considering Spatiotemporal Dependencies

DyGCN-LSTM: A dynamic GCN-LSTM based encoder-decoder framework for multistep traffic prediction

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

Spatial-temporal gated graph convolutional network: a new deep learning framework for long-term traffic speed forecasting

Foresee Urban Sparse Traffic Accidents: A Spatiotemporal Multi-Granularity Perspective

Deep Learning for Short-Term Traffic Flow Prediction

Similarity-based Feature Extraction for Large-scale Sparse Traffic Forecasting

Revisiting Spatial-Temporal Similarity: A Deep Learning Framework for Traffic Prediction

A Spatial-Temporal Framework Including Traffic Diffusion for Short-Term Traffic Prediction.

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