Reinforcement Learning Technique in Multiple Motorway Access Control Strategy Design
AbstractAn appropriately designed motorway access control can decrease the total travel time spent in the system up to 30% and consequently increase the merging operations safety. To date, implemented traffic responsive motorway access control systems have been of local or regulatory type and not truly adaptive in the real sense of the meaning. Hence, traffic flow can be influenced positively by numerous intelligent transportation system (ITS) techniques. In this paper a contemporary approach is presented. It considers the design philosophy of an optimal and adaptive closed-loop multiple motorway access control strategy. The methodology proposed uses the artificial intelligence technique - known as reinforcement learning (RL) with multiple agents, and applies the Q-learning algorithm. One segment of the motorway network with three lanes in each direction and three motorway entries was designed. The detectors and traffic signals were placed at the entries (ramps). Traffic flows and traffic occupancy on the main line as well as the traffic demand on the motorway entries were taken as input model variables. The output variables referred to the travel speed on the corridor, the total travel time, and the total stop time. VISSIM micro-simulator and direct programming of the simulator functions were used in order to implement the RL technique. The peak hour was chosen for the time of simulation. The model was tested in two phases. Its effectiveness was compared to ALINEA. It was observed that the proposed strategy was capable of responding both to dynamic sensory inputs from the environment and to dynamically changing environment. The model of the environment and supervision were not required. The control policy changed as response to the inherent system characteristic changes. It was confirmed that the strategy was truly adaptive and real-time responsive to the traffic demand on the corridor. KEY WORDS: motorway access, traffic flows, control, strategy, artificial intelligence, Q-Learning, simulation
Authors who publish with this journal agree to the following terms:
- Authors retain copyright and grant the journal right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgement of the work's authorship and initial publication in this journal.
- Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the journal's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgement of its initial publication in this journal.
- Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See The Effect of Open Access).