An official website of the United States government
This study proposes a data fusion and deep learning (DL) framework that learns high-level traffic features from network-level images to predict large-scale, multi-route, speed and volume of connected vehicles (CVs). We present a scalable and parallel method of processing statewide CVs’ trajectory data that leads to real-time insights on the micro-scale in time and space (two-dimensional (2D) arrays) on graphics processing unit (GPUs) using the Nvidia rapids framework and dask parallel cluster, which provided a 50× speed-up in the data extraction, transform and load (ETL). A UNet model is then applied to perform feature extraction and multi-route speed and volume channels over a multi-step prediction horizon. The accuracy and robustness of the proposed model are evaluated by taking different road types, times of day and image snippets and comparing the model to benchmarks: Convolutional Long–Short-Term Memory (ConvLSTM) and a historical average (HA). The results show that the proposed model outperforms benchmarks with an average improvement of 15% over ConvLSTM and 65% over the HA. Comparing the image snippets from each prediction model to the actual image shows that image textures were highly similar in UNet to the benchmark models used. UNet’s dominance in performing image predictions was also evident in multi-step forecasting, where the increase in errors was relatively minimal over longer prediction horizons.
more »
« less
Probabilistic Traffic State Prediction Based on Vehicle Trajectory Data
Accurate prediction of traffic flow dynamics is a key step towards effective congestion mitigation strategies. The dynamic nature of traffic flow and lack of comprehensive data coverage (e.g., availability of data at loop detector locations), however, have historically prevented accurate traffic state prediction, leading to the widespread utilization of reactive congestion mitigation strategies. The introduction of connected automated vehicles provides an opportunity to address this challenge. These vehicles rely on trajectory-level prediction of their surrounding traffic environment to plan a safe and efficient path. This study proposes a methodology to utilize the outcome of such predictions to estimate the future traffic state. Moreover, the same approach can be applied to data from connected vehicles for traffic state prediction. Since in many driving scenarios, more than one maneuver is feasible, it is more logical to predict the location of the vehicles in a probabilistic manner based on the probability of different maneuvers. The key contribution of this study is to introduce a methodology to convert such probabilistic trajectory predictions to aggregate traffic state predictions (i.e., flow, space–mean speed, and density). The key advantage of this approach (over directly predicting traffic state based on aggregated traffic data) is its ability to capture the interactions among vehicles to increase the accuracy of the prediction. The down side of this approach, on the other hand, is that any increase in the prediction horizon reduces the accuracy of prediction (due to the uncertainty in the vehicles’ interactions and the increase in the possibility of different maneuvers). At the microscopic level, this study proposes a probability based version of the time–space diagram, and at the macroscopic level, this study proposes probabilistic estimates of flow, density, and space–mean speed using the trajectory-level predictions. To evaluate the effectiveness of the proposed approach in predicting traffic state, the mean absolute percentage error for each probabilistic macroscopic estimate is evaluated on multiple subsamples of the NGSIM US-101 and I-80 data sets. Moreover, while introducing this novel traffic state prediction approach, this study shows that the fundamental relation among the average traffic flow, density, and space–mean speed is still valid under the probabilistic formulations of this study.
more »
« less
- Award ID(s):
- 2112650
- PAR ID:
- 10592413
- Publisher / Repository:
- Springer
- Date Published:
- Journal Name:
- Data Science for Transportation
- Volume:
- 5
- Issue:
- 3
- ISSN:
- 2948-135X
- Subject(s) / Keyword(s):
- Probabilistic traffic prediction · Fundamental diagram · Vehicle trajectory data
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Traffic congestion anomaly detection is of paramount importance in intelligent traffic systems. The goals of transportation agencies are two-fold: to monitor the general traffic conditions in the area of interest and to locate road segments under abnormal congestion states. Modeling congestion patterns can achieve these goals for citywide roadways, which amounts to learning the distribution of multivariate time series (MTS). However, existing works are either not scalable or unable to capture the spatial-temporal information in MTS simultaneously. To this end, we propose a principled and comprehensive framework consisting of a data-driven generative approach that can perform tractable density estimation for detecting traffic anomalies. Our approach first clusters segments in the feature space and then uses conditional normalizing flow to identify anomalous temporal snapshots at the cluster level in an unsupervised setting. Then, we identify anomalies at the segment level by using a kernel density estimator on the anomalous cluster. Extensive experiments on synthetic datasets show that our approach significantly outperforms several state-of-the-art congestion anomaly detection and diagnosis methods in terms of Recall and F1-Score. We also use the generative model to sample labeled data, which can train classifiers in a supervised setting, alleviating the lack of labeled data for anomaly detection in sparse settings.more » « less
-
Traffic congestion anomaly detection is of paramount importance in intelligent traffic systems. The goals of transportation agencies are two-fold: to monitor the general traffic conditions in the area of interest and to locate road segments under abnormal congestion states. Modeling congestion patterns can achieve these goals for citywide roadways, which amounts to learning the distribution of multivariate time series (MTS). However, existing works are either not scalable or unable to capture the spatial-temporal information in MTS simultaneously. To this end, we propose a principled and comprehensive framework consisting of a data-driven generative approach that can perform tractable density estimation for detecting traffic anomalies. Our approach first clusters segments in the feature space and then uses conditional normalizing flow to identify anomalous temporal snapshots at the cluster level in an unsupervised setting. Then, we identify anomalies at the segment level by using a kernel density estimator on the anomalous cluster. Extensive experiments on synthetic datasets show that our approach significantly outperforms several state-of-the-art congestion anomaly detection and diagnosis methods in terms of Recall and F1-Score. We also use the generative model to sample labeled data, which can train classifiers in a supervised setting, alleviating the lack of labeled data for anomaly detection in sparse settings.more » « less
-
Car-following (CF) behavior is a fundamental of traffic flow modeling; it can be used for the virtual testing of connected and automated vehicles and the simulation of various types of traffic flow, such as free flow and traffic oscillation. Although existing CF models can replicate the free flow well, they are incapable of simulating complicated traffic oscillation, and it is difficult to strike a balance between accuracy and efficiency. This article investigates the error variation when the traffic oscillation is simulated by the intelligent driver model (IDM). Then, it divides the traffic oscillation into four phases (coasting, deceleration, acceleration, and stationary) by using the space headway of multiple steps. To simulate traffic oscillation between multiple human-driven vehicles, a dynamic transformation CF model is proposed, which includes the long-time prediction submodel [modified sequence-to-sequence (Seq2seq)] model, short-time prediction submodel (Transformer), and their dynamic transformation strategy]. The first submodel is utilized to simulate the coasting and stationary phases, while the second submodel is utilized to simulate the acceleration and deceleration phases. The results of experiments indicated that compared to K -nearest neighbors, IDM, and Seq2seq CF models, the dynamic transformation CF model reduces the trajectory error by 60.79–66.69% in microscopic traffic flow simulations, 7.71–29.91% in mesoscopic traffic flow simulations, and 1.59–18.26% in macroscopic traffic flow simulations. Moreover, the runtime of the dynamic transformation CF model (Inference) decreased by 14.43–66.17% when simulating the large-scale traffic flow.more » « less
-
Traffic energy consumption estimation is significant for the sustainable transportation. However, it is difficult to directly employ macro traffic flow data to accurately estimate the traffic energy consumption due to many traffic energy consumption models need second-by-second vehicle trajectory. To solve this problem, this paper proposes a traffic energy consumption model based on the macro-micro data, which the macro data derived from the fixed-location sensors and sparse micro data derived from the Connected and Automated Vehicles (CAVs). The completed vehicle trajectories are constructed by the nonparametric kernel smoothing algorithm and variational theory. To test the performance of the proposed method, the Next Generation Simulation micro (NGSIM) dataset and Caltrans Performance Measurement System macro dataset obtained from the same road and time are used. The results indicate that the proposed method not only can reflect the characteristics of traffic kinematic waves caused by traffic congestion, but also minimize the errors generated by the macro-micro transformation. In addition, it can significantly improve the accuracy of energy consumption estimation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1007/s42421-023-00085-8
