iTensor: intelligent traffic prediction and controls for smart city
The fast ICT development has stimulated the innovations and applications in intelligent vehicle and transport systems technologies as well as associated industrial development. For management and operation of road transport, there is big potential to make the system more intelligent, energy-efficient and eco-friendly. During the past decade, we have been developing Intelligent Transportation Systems (ITS) solutions for modern traffic information and management systems in the context of smart city.
The first phase of iTensor has developed an agent-based traffic control framework using model-free and model based reinforcement learning. The methodological approaches have large potentials to be applied with various traffic signal control strategies. For example, our original developments were to improve traffic signal controls in Sweden, where group-based traffic signals are applied in engineering practice. The work can be referred to the following papers:
J. Jin and X. Ma, "A group-based traffic signal control with adaptive learning ability," Engineering Applications of Artificial Intelligence, vol. 65, pp. 282-293, 2017.
J. Jin and X. Ma, "Hierarchical multi-agent control of traffic lights based on collective learning," Engineering Applications of Artificial Intelligence, vol. 68, pp. 236-248, 2018.
Moreover, the RL-based platform has been generalized for the stage-based signal control as well as multi-objective policy goals. A multi-objective reinforcement learning approach has been developed for the learning-based control approaches. Relevant development can be seen in the papers such as
J. Jin and X. Ma, "A Multi-Objective Agent-Based Control Approach With Application in Intelligent Traffic Signal System," IEEE Transactions on Intelligent Transportation Systems, Vol. 20, No. 10, pp. 3900-3912, 2019.
In order to implement intelligent traffic controls, accurate and real-time traffic estimation and prediction are important for the ITS system. As increasing transport data from various sources has a direct impact on traffic prediction, data-driven approach, supported by the state of the art machine learning techniques, has become our main research direction in iTensor. The research can be referred to the following papers:
J. Jin and X. Ma, "A non-parametric Bayesian framework for traffic-state estimation at signalized intersections," Information Sciences, vol. 498, pp. 21-40, 2019.
M. Sederlin, X. Ma and J. Jin, "A Hybrid Modelling Approach for Traffic State Estimation at Signalized Intersections", The 24th IEEE Conference on Intelligent Transportation Systems, 2021.
The methodological approaches for traffic prediction and control have been extended and implemented in engineering solutions in collaboration with industrial partners. For example, a recent work extending from early study is an intelligent recommender model for human-in-the-loop traffic control using machine learning kernals, with a detailed application in the city of Hangzhou:
J. Jin, D. Rong, F. Zhu, H. Guo, X. Ma, and F. Wang, "PRECOM: A Parallel Recommendation Engine for Control, Operations, and Management on Congested Urban Traffic Networks" , IEEE Transactions on Intelligent Transportation Systems, accepted in 2021.
In the second phase of iTensor, we aim to further develop the reinforcement learning based methods and advanced computational approaches for implementing large-scale intelligent traffic signal control problems. Moreover, we are modifying the iTensor computational platform for more efficient learning and various traffic control problems. In addition to control development, intelligent traffic prediction for large-scale network using the state of art machine learning models and high-performance computing becomes an essential part of the ITS application and our reseach direction.