科研内容 Research

我组科研重点强调控制、优化理论和智能网联系统应用的有机结合。主要理论成果包括随机非线性流体网络的分析和控制、非传统排队网络的分析及控制。主要方法是模型预测控制和强化学习算法的结合、动态系统安全博弈分析等，在世界范围内具有开创性、前沿性。主要应用领域包括混合自主性条件下的队列行驶、无信号灯路口的最优/容错控制、信息物理扰动下的交通网络控制、智能网联运输系统的安全控制等。
We are interested in control and optimization problems in smart and connected systems. We focus on the theory of nonlinear flow network control, non-conventional queuing network control. We synthesize model-predictive and learning-based methods to design control algorithms that are tolerant of random faults and secure against strategic attacks. Specific applications include vehicle platooning in mixed autonomy, resilient control of high-speed signal-free intersections, traffic network control under cyber-physical disruptions, and secure control of smart and connected transportation systems.

我组长期和以下高校合作 We collaborate with researchers from the following institutes：

我组科研活动（正在或曾经）受以下单位资助 Our research is or was sponsored by：

基于强化学习的网络控制方法及其安全防护策略
Learning-Based Network Control and Its Security Theory

本项目首先通过联立李雅普诺夫函数和近似价值函数，理解状态演化-参数迭代的双重收敛关系；再基于分段决定马氏过程理论，将对抗博弈融入算法训练及执行过程的收敛性分析，以辨识算法的失效模式并探讨相应防护手段；最后探究未知环境中的信息筛选与反推原理，在训练、执行阶段分别引入可控对抗数据、在线更新机制，实现算法在未知环境中的“全周期”防护，并完成跨场景多模式仿真验证。
This project builds on the theory of stochastic controloptimizationgames, studies the joint convergence of network state and learning parameters, characterizes the failure mechanisms of learning-based control, and explores the principles for detecting attacks from unknown sources. This project constructs a solid foundation for the security theory of learning-based network control, explores reliable and efficient defense methods, and validates in simulations of typical scenarios.
资助单位：国家自然科学基金委员会
Sponsor: National Natural Science Foundation of China

车联网的韧性控制与优化
Resilient Control and Optimization of Internet of Vehicles

我们重点关注车联网技术应用过程中面临的两大挑战——信息安全和人为因素。针对信息安全隐患，提出量化其风险的理论框架，并结合自动控制、强化学习等方法，给出实际操作的安全控制/优化算法。针对人为因素，引入博弈模型，分析人为决策对车联网系统演化的影响，并设计相应决策算法。
We focus on two key challenges involved in the application of the technology of Internet of Vehicles (IoV), viz. cyber security and human factors. For the former, we propose a framework to quantify the impact of cyber security risks. We use ideas from automatic control and reinforcement learning to develop implementable secure control/optimization algorithms. For the latter, we introduce game-theoretic models to analyze the impact of human decisions on the evolution of IoVs and develop corresponding decision-making algorithms
资助单位：国家自然科学基金委员会
Sponsor: National Natural Science Foundation of China

动态流网络上的安全博弈 Security games on dynamic flow networks

我们关注动态流网络上的一系列安全问题。该类问题可由确定性或随机性博弈模型刻画，博弈双方为恶意对手（攻击方）和系统运营者（防守方）。问题的关键是处理上述博弈对动态流网络稳定性和效率（吞吐量、时延）的影响。
We consider a series of security problems over dynamic flow networks. The security problems are characterized by deterministic or stochastic games between a malicious adversary (attacker) and a system operator (defender). The key is to capture the impact of such games on the stability and efficiency (throughput and delay) of the dynamic flow networks.
资助单位：上海交通大学密西根学院
Sponsor: Shanghai Jiao Tong University UM Joint Institute

智能网联汽车队列行驶宏观机理及自适应控制方法研究
Adaptive control for connected and autonomous vehicle platooning

我们首先提出了原创的混合排队模型，解决了传统车流模型无法描述路基系统和CAV队列间点对点互动机制的问题，同时解决了已有CAV运动学模型难以推导系统宏观性能（通行效率、等待时间）的问题。基于此模型，我们综合运用经典排队论、马式过程稳定性理论、单调系统控制理论，定量分析了CAV队列对系统性能的影响，给出了混行情境下理论最优控制策略，并在复杂仿真平台得到了验证。我们还将该模型混合状态变量的思想推广到交通流的偏微分方程模型，得到了细化的控制算法。
We propose an original hybrid queuing model for the control of CAV platoons in mixed autonomy. This model reconciles the lack of scalability and tractability of microscopic, trajectory-based models and the lack of point-to-point tracking and intervention of macroscopic, flow-based models. We use ideas from queuing theory and Markov decision processes to design platooning operations that are adaptive to non-CAV traffic flows as well as environmental non-stationarities.
资助单位：国家自然科学基金委员会
Sponsor: National Natural Science Foundation of China

基于学习的容错交通管理算法
Learning-Based Fault-Tolerant Traffic Management Algorithms for Intelligent Transportation Systems

该课题研究交通事故及数据损坏等扰动下的智能公路交通管理方法。我们提出了一套定量分析随机扰动导致的通行能力损失、设计缓解扰动影响的韧性车流管控方法。运用马式链模型来描述公路通行能力的动态浮动，并融合该模型和经典交通流模型，得到了可追踪车流时空演化的模型。提出了抗扰动的匝道控制方法，并给出了该方法所能达到性能的理论上下界。
This project studies traffic management in the face of random disruptions including traffic incidents and imperfect data. We used stochastic models to capture the capacity loss due to random disruptions. We integrated the disruption models with classical traffic flow models to develop ramp metering strategies that are resilient against disruptions.
资助单位：美国国家科学基金会
Sponsor: United States National Science Foundatio