Machine learning and fuzzy cognitive maps in a hybrid approach toward freeway on-ramp traffic control

The infrequent emergence of traffic congestion on freeways can result in the decline of the transportation system over time. Without the implementation of appropriate countermeasures, congestion can escalate, leading to unfavorable impacts on other aspects of the traffic network. As a result, there is a greater need for reliable and optimal traffic control. The goal of this research is to manage the number of vehicles entering the main freeway from the ramp merging area, in order to balance the demand and capacity to satisfy the maximum utilization of the freeway capacity. Despite extensive research into different ramp metering techniques, this study aims to utilize the fuzzy cognitive map as a macroscopic traffic flow model in conjunction with the Q-learning algorithm. This combination prevents freeway congestion and maintains optimal performance by keeping freeway density below a key threshold. The inherent uncertainty of traffic conditions is addressed through the application of reinforcement learning, which is constructed on the principles of the Markov decision process. This approach represents an exploration-exploitation trade-off, as implemented through the Q-learning algorithm. The proposed technique was evaluated for its efficacy in the regulation of freeway ramp metering in both controlled and uncontrolled simulations. The findings demonstrate a significant improvement in the control of the mainstream traffic flow.