State-of-the-practice traffic management relies on the full knowledge of the covered area, and needs permanent maintenance and monitoring to ensure that the system reacts correctly to each change in the infrastructure, in the coverage, on in the demand trends/patterns and modes.
The traffic process is by nature a multi-agent / multi-system ecosystem where each constituent system aims for an identified, generally short-term, purpose. The overall satisfaction of the traffic process is a technically sophisticated challenge requiring an ad-hoc fine-tuning and calibration approach. Therefore, a methodology where the different agents are given a certain freedom in deciding for local decisions but also where a central element influences regularly the local decisions per the latest full knowledge of the ecosystem’s performance indicators, is worth to be investigated and tested in a realistic context.
Control strategy description:
The technical system of systems in our case is the traffic process. The constituent systems are divided into two classes: I signalized intersections and II cooperative vehicles. Each constituent system is taking local and short decisions independently in reaction to its close environment. The figure below illustrates the concept of traffic system of systems.
In the case of signalized intersections, the environment conditions are the traffic demand in the different intersection arms measured in terms of occupancies/queues by the magnetic loop detectors next to each traffic light. The decisions made by this local controller are the signal programs applied in the next cycle.
In the case of cooperative vehicles, the environment conditions are the traffic light signalling in the next intersection in the trip, and the queue length next to it. The decisions made by this local controller is the optimal cruise speed to cross during the green light and to minimize stops and useless acceleration.
The global optimizer (L4G optimization tool – L4GCAO) learns about the current traffic situation from the local controllers of the first class, assuming that they are interacting with the whole vehicle fleets not only the cooperative ones. It generates however optimization parameter values for both classes of constituent systems aiming to reduce queuing next to intersections and maximize the flowing, allowing a maximum of vehicles to use the network in a shorter time. This has also the impact of reducing fuel consumption and emissions per vehicle and at the neighbourhood of each intersection.
Reference/Base Case control:
The proposed test bed is a road stretch of 7 intersections in the road of Munich, Germany. The infrastructure is operated under the command of the South Bavarian Road Administration by the private operator Siemens. It has little actuation and runs mostly in a scheduled fixed time control, i.e. one signal plan is repeating constantly during the allocated time of the day. This fixed control has been actually designed and calibrated in previous research projects focusing on traffic optimization, with the aim to obtain a fluent green wave along the 7 intersections, and it produced a good performance.
As for the second class of constituent systems, there are currently in practice no cooperative vehicles running on the network in a usual basis. In the same previous research project, there was however a concept for a speed recommendation smartphone app based on the known signal programs prepared jointly. This speed recommendation application gave as well satisfactory results.
Implementation architecture setup:
The test bed conditions required an ad-hoc architecture to fit to the legal and institutional requirements. All traffic control related information transits necessarily through the systems of the road authority’s contractor (Siemens) in both directions (from/to the field). On the other hand, an already established client/server architecture for the smartphone apps equipping the connected vehicles was in place and hosted by the TUM partner. Therefore, the final concept to integrate the Local4Global TSoS was made of 3 central locations besides the field’s constituent systems: one for direct communication with each category of constituent systems and one L4G location to host the optimization service, in the premises of the TEKNIKER-IK4 partner.
Another requirement of the field conditions was the impossibility to host the local control algorithm directly in the traffic light controller cabinet in the case of systems of class I. Therefore the local control of intersections was ported to the central L4G location but to work on a strictly local basis, i.e. receives only local information of the intersections and sends back local signaling decisions. Besides, freshly generated signal programs were avoided to eliminate any risk of safety related constraint such as the clearing time verification. This was surmounted through the adoption of a predefined signal plan library for each intersections that are safety checked. The local control must then just pick the closest option to the optimal generated signal plan.
Several criteria were suggested to quantify the system of systems’ objective satisfaction. The fluency of traffic at intersections is described through the number of stops, waiting time, queue lengths, percentage of crosses without stops. The overall traffic process performance is quantified through the average network mean speed. This metric allows a spatially and temporally independent insight that reflects directly the adaptability of the implemented process management strategy.
Application Test Scenario:
The Local4Global system in Spain was connected to the road traffic computer in Munich to collect field data during 4 weeks in June 2016 on regular weekdays.
Each day datasets corresponding to a daily off-peak period between morning and evening and another to the evening peak period were injected separately and consecutively in the Local4Global model that called the signalling local controllers generating the signalling decisions, commuting them to signal controllers in the model and to connected vehicles in the model. Each simulation is iterated 10 times with the global optimizer turned on for fine-tuning.
Contingency to field conditions such as detector failures were tested during this scenario.
The connected simulation allowed to extract realistic results that can be compared to the reference scenario in the same demand and boundary conditions. L4G showed a capacity to accommodate a greater throughput of traffic than the baseline. The comparison shows also a good improvement (42%) in the case of peak hour traffic and a very slight improvement (1%) during off-peak traffic, in terms of average network speed. Between the L4G scenarios of with and without C2X co-operative vehicles, it was shown that the L4G system with 10% C2X co-operative vehicles generally perform similarly to the L4G system without such vehicles.
adminResults of the Cooperative Traffic TSoS Use Case