Short-term traffic speed prediction and forecasting using machine learning analysis of spatiotemporal traffic speed dependencies in probe and weather data
A framework for modeling traffic speed in a transportation network analyzes both the spatial and temporal dependencies in probe-based traffic speeds, historical weather data, and forecasted weather data, using multiple machine learning models. A decentralized partial least squares (PLS) regression model predicts short-term speed using localized, historical probe-based traffic data, and a deep learning model applies the predicted short-term speed to further estimate traffic speed at specified times and at specific locations in the transportation network for predicting traffic bottlenecks and other future traffic states
1. A method of characterizing traffic congestion in a transportation network, comprising:
receiving, as input data, probe data that includes localized, historical link-based speed information in a transportation network, and vehicle data representing an operation of one or more autonomous vehicles within the transportation network;
analyzing at least the probe data in a plurality of data processing elements within a computing environment that includes one or more processors and at least one computer-readable non-transitory storage medium having program instructions stored therein which, when executed by the one or more processors, cause the one or more processors to execute the plurality of data processing elements to characterize spatio-temporal dependencies in traffic speed, by:
selecting one or more predictors from features in the probe data relevant to each link in the transportation network,
constructing a matrix defined by a number of time steps, and a plurality of features representing temporal characteristics in the one or more predictors selected for each link,
generating a multi-variate time-series dataset representing a predicted short-term traffic speed for each link from the matrix,
transforming differential time sequences in the multi-variate time-series dataset in a multi-layered neural network having a sequence-to-sequence architecture to generate an estimate of traffic speed at one or more specific locations of the transportation network at one or more specified times where the one or more autonomous vehicles are operating derived from the vehicle data; and
predicting traffic bottlenecks at the one or more specific locations of the transportation network where the one or more autonomous vehicles are operating at the one or more specified times from the estimate of the traffic speed; and
remotely adjusting the operation of the one or more autonomous vehicles within the transportation network.
2. The method of claim 1 , wherein the one or more autonomous vehicles include remotely-piloted passenger vehicles and remotely-piloted commercial vehicles.
3. The method of claim 1 , wherein the input data further includes incident data that includes roadway maintenance activity within the transportation network, and wherein the analyzing at least the probe data includes analyzing the incident data.
4. The method of claim 1 , wherein the input data further includes weather data that includes one or more of current weather data, historical weather conditions, near-term forecasts of weather conditions, extended-range forecasts of weather conditions, and predicted weather conditions for the geographical area that includes the transportation network, and wherein the analyzing at least the probe data includes analyzing the weather data.
5. The method of claim 1 , wherein the vehicle data includes routing information for the one or more autonomous vehicles, and wherein the operation of the one or more autonomous vehicles includes operation along a recommended route.
6. The method of claim 1 , wherein the sequence-to-sequence architecture is comprised of an encoder generating a hidden neural network state representing the multivariate time-series dataset, a decoder generating an output sequence, and a fully-connected layer.
7. The method of claim 1 , further comprising applying an activation function at an output of the multi-layered neural network to derive the estimate of the traffic speed at the one or more locations.
8. The method of claim 1 , wherein the matrix maps the localized, historical link-based speed information to a future traffic state for each link.
9. The method of claim 1 , wherein the traffic bottlenecks represent predicted delays characterized by a reduction in traffic speed across a specific distance comprised of one or more links.
10. The method of claim 1 , further comprising generating a forecast of a future traffic state from the estimate of traffic speed to predict the traffic bottlenecks, wherein the forecast of the future traffic state includes a forecast of one or more of volume, flow, and travel time for the one or more specific locations at one or more specified times.
11. The method of claim 1 , further comprising generating visualizations of one or more of the traffic bottlenecks, an adjusted operation of the one or more autonomous vehicles, and the recommended route for the one or more autonomous vehicles, for a graphical user interface.
12. The method of claim 1 , wherein the one or more predictors include traffic speed features representing link-based speed data from every link, and temporal features representing a day, hour, and minute of each data point in the link-based speed data.
13. The method of claim 12 , wherein the link-based speed data is segmented based on variances in the transportation network, the variances including changes in geometry, intersections, changes in posted speed limit, and lane configuration.
14. The method of claim 1 , wherein one or more traffic signals are adjusted for routing of the one or more autonomous vehicles.
15. A method, comprising:
modeling spatio-temporal dependencies in traffic speed from input data comprised of probe data that includes localized, historical link-based speed information in a transportation network, and vehicle data representing an operation of one or more autonomous vehicles within the transportation network, by:
selecting one or more predictors from features in the probe data relevant to each link in the transportation network,
predicting short-term traffic speed for each link, by constructing a matrix defined by a number of time steps, and a plurality of features representing temporal characteristics in the one or more predictors selected for each link, and generating a multi-variate time series dataset for each link from the matrix,
estimating a traffic speed at one or more locations of the transportation network at one or more specified times where the one or more autonomous vehicles are operating, by applying the multi-variate time-series dataset to a multi-layered neural network to transform differential time sequences in the multi-variate time-series dataset; and
analyzing an estimate of the traffic speed at the one or more locations of the transportation network to predict traffic bottlenecks where the one or more autonomous vehicles are operating at the one or more specified times, the traffic bottlenecks representing predicted delays characterized by a reduction in traffic speed,
wherein the operation of the one or more autonomous vehicles is remotely controlled within the transportation network.
16. The method of claim 15 , wherein the one or more autonomous vehicles include remotely-piloted passenger vehicles and remotely-piloted commercial vehicles.
17. The method of claim 15 , wherein the input data is further comprised of incident data that includes roadway maintenance activity within the transportation network.
18. The method of claim 15 , wherein the input data is further comprised of weather data that includes one or more of current weather data, historical weather conditions, near-term forecasts of weather conditions, extended-range forecasts of weather conditions, and predicted weather conditions for the geographical area that includes the transportation network.
19. The method of claim 15 , wherein the vehicle data includes routing information for the one or more autonomous vehicles, and wherein the operation of the one or more autonomous vehicles includes operation along a recommended route.
20. The method of claim 15 , wherein the multi-layered neural network has a sequence-to-sequence architecture comprised of an encoder generating a hidden neural network state representing the multivariate time-series data, a decoder generating an output sequence, and a fully-connected layer.
21. The method of claim 15 , further comprising applying an activation function at an output of the multi-layered neural network to derive the estimate of the traffic speed at the one or more locations.
22. The method of claim 15 , wherein the matrix maps the localized, historical link-based speed information to a future traffic state for each link.
23. The method of claim 15 , wherein the traffic bottlenecks represent predicted delays characterized by a reduction in traffic speed across a specific distance comprised of one or more links.
24. The method of claim 15 , further comprising generating a forecast of a future traffic state from the estimate of traffic speed to predict the traffic bottlenecks, wherein the forecast of the future traffic state includes a forecast of one or more of volume, flow, and travel time for the one or more specific locations at one or more specified times.
25. The method of claim 15 , further comprising generating visualizations of one or more of the traffic bottlenecks, an adjusted operation of the one or more autonomous vehicles, and the recommended route for the one or more autonomous vehicles, for a graphical user interface.
26. The method of claim 15 , wherein the one or more predictors include traffic speed features representing link-based speed data from every link, and temporal features representing a day, hour, and minute of each data point in the link-based speed data.
27. The method of claim 26 , wherein the link-based speed data is segmented based on variances in the transportation network, the variances including changes in geometry, intersections, changes in posted speed limit, and lane configuration.
28. The method of claim 15 , wherein one or more traffic signals are adjusted for routing of the one or more autonomous vehicles.
29. A system for characterizing traffic congestion in a transportation network, comprising:
a data collection element configured to receive input data comprised of probe data that includes localized, historical link-based speed information in a transportation network, and vehicle data representing an operation of one or more autonomous vehicles within the transportation network; and
one or more machine learning models, configured to analyze at least the probe data to characterize spatio-temporal dependencies in traffic speed, the one or more machine learning models including:
a partial least squares regression model configured to predict short-term traffic speed for each link in the transportation network, by
selecting one or more predictors from features in the input data relevant to each link in the transportation network, and
constructing a matrix defined by a number of time steps, and a plurality of features representing temporal characteristics in the one or more predictors selected for each link, and generating a multi-variate time series dataset for each link from the matrix; and
a deep learning model configured to transform differential time sequences in the multi-variate time-series dataset and estimate a traffic speed at one or more locations of the transportation network at one or more specified times where the one or more autonomous vehicles are operating, the deep learning model comprised of a multi-layered neural network having a sequence-to-sequence architecture,
wherein an estimate of the traffic speed at the one or more locations of the transportation network is analyzed to predict traffic bottlenecks where the one or more autonomous vehicles are operating at the one or more specified times, the traffic bottlenecks representing predicted delays characterized by a reduction in traffic speed, and
wherein the operation of the one or more autonomous vehicles is remotely controlled within the transportation network.
30. The system of claim 29 , wherein the one or more autonomous vehicles include remotely-piloted passenger vehicles and remotely-piloted commercial vehicles.
31. The system of claim 29 , wherein the input data further includes incident data that includes roadway maintenance activity within the transportation network, and wherein the one or more machine learning models are further configured to the incident data.
32. The system of claim 29 , wherein the input data further includes weather data that includes one or more of current weather data, historical weather conditions, near-term forecasts of weather conditions, extended-range forecasts of weather conditions, and predicted weather conditions for the geographical area that includes the transportation network, and wherein the one or more machine learning models are further configured to analyze the weather data.
33. The system of claim 29 , wherein the vehicle data includes routing information for the one or more autonomous vehicles, and wherein the operation of the one or more autonomous vehicles includes operation along a recommended route.
34. The system of claim 29 , wherein the sequence-to-sequence architecture is comprised of an encoder generating a hidden neural network state representing the multivariate time-series dataset, a decoder generating an output sequence, and a fully-connected layer.
35. The system of claim 29 , wherein an activation function at an output of the multi-layered neural network derives the estimate of the traffic speed at the one or more locations.
36. The system of claim 29 , wherein the matrix maps the localized, historical link-based speed information to a future traffic state for each link.
37. The system of claim 29 , wherein the traffic bottlenecks represent predicted delays characterized by a reduction in traffic speed across a specific distance comprised of one or more links.
38. The system of claim 29 , wherein a forecast of a future traffic state is generated from the estimate of traffic speed to predict the traffic bottlenecks, and wherein the forecast of the future traffic state includes a forecast of one or more of volume, flow, and travel time for the one or more specific locations at one or more specified times.
39. The system of claim 29 , wherein visualizations of one or more of the traffic bottlenecks, an adjusted operation of the one or more autonomous vehicles, and the recommended route for the one or more autonomous vehicles, are generated for a graphical user interface.
40. The system of claim 29 , wherein the one or more predictors include traffic speed features representing link-based speed data from every link, and temporal features representing a day, hour, and minute of each data point in the link-based speed data.
41. The system of claim 40 , wherein the link-based speed data is segmented based on variances in the transportation network, the variances including changes in geometry, intersections, changes in posted speed limit, and lane configuration.
42. The system of claim 29 , wherein one or more traffic signals are adjusted for routing of the one or more autonomous vehicles.