Spatiotemporal forecasting has various applications in neuroscience, climate and transportation domain. Traffic forecasting is one canonical example of such learning task. The task is challenging due to (1) complex spatial dependency on road networks, (2) non-linear temporal dynamics with changing road conditions and (3) inherent difficulty of long-term forecasting. To address these challenges, we propose to model the traffic flow as a diffusion process on a directed graph and introduce Diffusion Convolutional Recurrent Neural Network (DCRNN), a deep learning framework for traffic forecasting that incorporates both spatial and temporal dependency in the traffic flow. Specifically, DCRNN captures the spatial dependency using bidirectional random walks on the graph, and the temporal dependency using the encoder-decoder architecture with scheduled sampling. We evaluate the framework on two real-world large scale road network traffic datasets and observe consistent improvement of 12% - 15% over state-of-the-art baselines.

/pdf/diffusion-convolutional-recurrent-neural-network-data-driven-3wefkkjq0q.pdf

Diffusion Convolutional Recurrent Neural Network: Data-Driven Traffic Forecasting

Transportation Networks. Optimality. Cost Functions. Deterministic User Equilibrium Assignment. Stochastic User Equilibrium Assignment. Trip Table Estimation. Network Reliability. Network Design. Conclusions. References. Index.

Transportation network analysis

The k-nearest neighbor (KNN) model is an effective statistical model applied in short-term traffic forecasting that can provide reliable data to guide travelers. This study proposes an improved KNN model to enhance forecasting accuracy based on spatiotemporal correlation and to achieve multistep forecasting. The physical distances among road segments are replaced with equivalent distances, which are defined by the static and dynamic data collected from real road networks. The traffic state of a road segment is described by a spatiotemporal state matrix instead of only a time series as in the original KNN model. The nearest neighbors are selected according to the Gaussian weighted Euclidean distance, which adjusts the influences of time and space factors on spatiotemporal state matrices. The forecasting accuracies of the improved KNN and of four other models are compared, and experimental results indicate that the improved KNN model is more appropriate for short-term traffic multistep forecasting than the other models are. This study also discusses the application of the improved KNN model in a time-varying traffic state.

A spatiotemporal correlative k-nearest neighbor model for short-term traffic multistep forecasting

Due to its paramount relevance in transport planning and logistics, road traffic forecasting has been a subject of active research within the engineering community for more than 40 years. In the beginning most approaches relied on autoregressive models and other analysis methods suited for time series data. More recently, the development of new technology, platforms and techniques for massive data processing under the Big Data umbrella, the availability of data from multiple sources fostered by the Open Data philosophy and an ever-growing need of decision makers for accurate traffic predictions have shifted the spotlight to data-driven procedures. This paper aims to summarize the efforts made to date in previous related surveys towards extracting the main comparing criteria and challenges in this field. A review of the latest technical achievements in this field is also provided, along with an insightful update of the main technical challenges that remain unsolved. The ultimate goal of this work is to set an updated, thorough, rigorous compilation of prior literature around traffic prediction models so as to motivate and guide future research on this vibrant field.

Road Traffic Forecasting: Recent Advances and New Challenges

Survey on traffic prediction in smart cities

Urban expressways usually experience several levels of service (LOS) because of the stop-and-go traffic flow caused by congestion. Moreover, multiple shock waves generate at different LOS interfaces. The dynamic of shock waves strongly influences the travel time reliability (TTR) of urban expressways. This study proposes a path TTR model that considers the dynamic of shock waves by using probability-based method to characterize the TTR of urban expressways with shock waves. Two model parameters are estimated, namely distribution of travel time (TT) per unit distance and travel distances in different LOS segments. Generalized extreme value distribution and generalized Pareto distribution are derived as distributions of TT per unit distance for six different LOS. Distribution parameters are estimated by using historical floating car data. Travel distances in different LOS segments are calculated based on shock wave theory. The range of TT along the path, which can help drivers arrange their trips, can be obtained from the TTR model. Finally, comparison is made among the proposed TTR model, generalized Pareto contrast model, which does not consider different LOS or existence of shock waves, and normal contrast model, which assumes TT per unit distance as normal distribution without considering shock wave. Results show that the proposed model achieves higher prediction accuracy and reduces the prediction range of TT. The conclusions can be further extended to TT prediction and assessment of measures to improve reliability of TT in a network.

A travel time reliability model of urban expressways with varying levels of service

Analysis of driving behaviors and related driver clustering is of great significance for improving driving safety, but traffic operation conditions (especially road types and operating speed) often are neglected in existing clustering studies, and the impact of excluding traffic conditions has not been investigated thoroughly. This research proposes an improved driver clustering framework by accounting for road types and average speed. The clustering results were compared with those without considering traffic conditions. The input data of more than 34 million records of second-by-second vehicle trajectories from 315 vehicles in Beijing were sliced into segments of 30 s, and these seconds were classified by road types (expressway versus non-expressway) and by 10-km/h average speed intervals. For each driver, the speed variation coefficients (SVCs), acceleration standard deviations (ASTDs), and average negative accelerations (ANAs) by traffic condition were entered into a Gaussian mixture model for an unsupervised clustering of drivers into types of prudent, normal, and aggressive drivers. The improved clustering framework is capable of capturing the variability of driving behaviors (especially dangerous driving behaviors such as sharp decelerations) across drivers, and the comparison demonstrated significant differences between the improved model and the original model with respect to the proportion of every driver type. The improved clustering framework performs better in both intraclass aggregation and interclass separation, and the results of this research indicate the need to consider traffic conditions in driving behavior–based clustering of drivers. 

Jianping Sun

Papers

A spatiotemporal correlative k-nearest neighbor model for short-term traffic multistep forecasting

A travel time reliability model of urban expressways with varying levels of service

Improved Driver Clustering Framework by Considering the Variability of Driving Behaviors across Traffic Operation Conditions