Heimdall RoadCast : Road Traffic Observation & Forecasting

Project awarded in response to call from Innovate UK, SRBI, Using-data-to-better-understand-and-respond-to-road-congestion (Phase 1), carried out as consultancy for Siemens Mobility Limited

Combining historical traffic data available in many regional traffic control system silos with real time information from connected traffic intersection controllers & contextual data from the region of deployment, Siemens and collaborators aim to produce the best quality urban traffic flow and congestion forecast available without requiring significant investment in infrastructure upgrades. This capability will provide triggers for real time congestion control strategies and interventions through existing traffic management systems such as Siemens STRATOS. By making modest infrastructure upgrades (e.g. a component change within a traffic intersection controller), congestion information based on vehicle class (car, truck, bus etc) may be derived on each intersection approach and the usefulness of this vehicle classification data will be investigated for use in more targeted strategies (locally in the traffic controller, or globally via the traffic management computer). An example intervention involves MOVA, a type of adaptive traffic control Investigate which traffic control modes and algorithms are likely to benefit most from vehicle type/class information. This work would be undertaken in conjunction with the traffic simulation team at Coventry University. Propose how a real time contextual data aggregator could be implemented and integrated into a traffic forecasting algorithm. This would be trialled in collaboration with Bristol City Council in phase 2. Phase 2 of the project aims to investigate the benefit cost ratio of each feasible idea and produce a minimum viable product to be trialled in Bristol.

The Aimsun traffic simulation was used to investigate, using microscopic simulation, the impact of data available (e.g. vehicle count, type and speed) on traffic. The specific use case investigated is the extension of the green time for slow moving vehicle based on the detection of these vehicles.
Aimsun lacks a MOVA interface, but it provides an adaptive signalised intersection based on (NCHRP Report 812, 2019). The latter, referred to as ‘actuated control plan’ was therefore used in all simulation studies. Four approaches were investigated and compared: fixed control plan, actuated control plan with green time extension irrespective of vehicle type, actuated control plan with pre-emption, actuated control plan with green time extension logic programmed in the application programming interface (API) using python. Three different versions were developed. The first extended the green time for trucks as long as they had specific detectors. The second extended the green time based on vehicle speed at the detector and the final one took into account the truck length to delay the detection of the truck until the rear of the vehicle had passed the detector. The last implementation replicated the behaviour of the loop based vehicle detection that requires the whole vehicle to have passed over the detector in order to classify it.
Two types of road networks were used. A realistic junction originally operated using MOVA and some simplified T junctions to design, implement and evaluate the algorithms as well as the impact of the slope of the road.
Two types of vehicles were simulated trucks and cars. There is scope to include more vehicle types as well as public transport. In this work the public transport lines were used to simulate equipped trucks given preference at the intersections.
The London 2017 fleet mix configuration was used to simulate the vehicle emissions. There is scope to implement specific model that take focuses on lorry fuel consumption at idle, for typical acceleration when starting from a stop position at a traffic light.
The saved simulations results included many criteria. This report focuses on travel time, emission and number of stops.
A good research platform has been developed combining Aimsun with python scripting to implement the algorithms. Further work should include linking with MATLAB and python to automate the simulation of alternative solutions and evaluate the use of optimisation within the traffic management algorithm.

Results
Results obtained during this initial phase of the work are preliminary and should be confirmed with a larger number of simulations. Overall it was found that extending green time could be beneficial and that this benefit depends on the junction geometry, complexity as well as the location of the detectors and the traffic flow. Due to time constraints the average of 10 simulations were used to evaluate the impact of each use case. This should be extended to 100 and include different weather conditions.

The results are summarised as follows:
• As expected, the slope of the road negatively affects the vehicle travel time and emissions by increasing the time required by lorries to reach the stop line and clear a junction. This results in the algorithm gaping out and the truck having to stop unnecessarily.
• The distance between the sensor used to detect the lorry and the stop lines affect the results. It is linked to the time given to extend the green. Simulation carried out with an extension of 4s led to improvement of more than 10% as the detector used for classification was moved away from the stop line. Note that these results are junction and traffic flow dependent. The time used to extend the green time should therefore the carefully calibrated for each junction and each phase.
• The tuning of the Aimsun actuated control plan had a significant impact on the performance of the simulation. The implementation of the green extension could make travel time worse on more instances than it could improve it. However, analysis of the results showed that the green extension was not activated as often as it should have been and that it could interfere with the built in ‘actuated’ plan that exploit the queue length on each link. It is recommended to develop a separate algorithm not based on some features of the existing actuated control plan to have full control of the algorithm behaviour.
• The methods used to adjust the green time extension included: fixed amount of time and speed dependent amount of time. The strategies investigated include extending the green time either by a constant amount or a speed dependent amount based on the slow vehicles detected by the road infrastructure. The simulation assumed loops were used, however the data provided by Aimsun are much richer and can include vehicle type and speed at the detector location. The algorithms were developed with the support of Aimsun Ltd. However, the budget did not allow significant involvement from Aimsun. It is recommended to involve the traffic simulation manufacturer at the onset of the next phase and allocate a more significant budget to enable academic staff to work closely with Aimsun Ltd.
Novelty of the work:
This research filled the gap between assuming a fully connected environment where all required vehicle and traffic light timing information is available and the current situation on most of the UK network where only vehicle presence is measurable with loops. Similarly to pre-emption, where giving priority to specific vehicles may negatively affect the overall traffic, favouring truck may have a similar effect of the overall traffic even if it benefits the lorries in terms of CO2 emissions. There is therefore an opportunity to further develop algorithm able to exploit vehicle information and integrate the decision to extend the time within the overall control algorithm.
Opportunities for future research
The following research opportunities have been identified based on the tools developed during this project:
1) MOVA carries out some optimisation to determine if the phases should be changed. The ‘Actuated’ plans implemented in Aimsun do not at this time use any optimisation. They are rule based. In the second phase of the work, it would be beneficial to modify the optimisation algorithm within MOVA to account for the presence of HGV whilst at the same time balancing the benefits of extending a green phase to enable to clear a slow moving vehicle with delaying the other phases. The impact of such optimisation could be studied by modifying aimsun actuated control plans to balance extending the green time for a number of truck (could be a single truck) by contrast to switching to another phase based on the queue length. Due to the required speed of response of these algorithms, it is recommended to optimise off-line for a range of situation and then derive a Pareto set for solutions storing all candidate solutions. A Pareto set contains solution that are optimal for all possible combination of importance factors between the different objectives. It means that it contains solutions that are best for emission as well as best for traffic flow or travel time. These optimal solutions, pre-calculated off-line, can then be used as a starting point to the online optimisation algorithm. The most important criteria (e.g. CO2, travel time) based on the situation at the time (e.g. peak time, off-peak, special incident, holidays) can then be used to select the most appropriate initial solution from the Pareto set. This solution can then be quickly refined based on the current flow and phase timing.
2) The current algorithm detects lorries on the links affected by the current phase. This means that if a lorry had been detected but did not have sufficient time to pass until the phase reaches its maximum green time, then the vehicle type information will not be used in the next occurrence of that phase. There is a need to extend this method to keep this information in memory for the next occurrence of this phase. Such information could be used to adapt the minimum green for the next occurrence of this phase as well as be exploited in the overall junction optimisation.
3) One of the algorithm implemented extend the green time based on the speed of the vehicles. This is similar to assuming that vehicles are connected and able to let the junction controller know about their speed. Such information could also be derived using cameras or a set of loops.
NCHRP Report 812, 2019, TRB’s National Cooperative Highway Research Program (NCHRP) Report 812: Signal Timing Manual - Second Edition, https://www.nap.edu/download/22097, Last accessed 15/3/2019.