由于不同智能化等级车辆的出现，越来越多的智能网联车辆（Connected Vehicles，CVs）与常规车辆（Regular Vehicles，RVs）混合行驶在道路上是未来交通发展的必经阶段。准确的交通流状态对确保安全高效通行尤为重要。目前，路侧已大规模部署了多传感器来感知道路交通环境。并且，道路中网联车的通信能力具有提供更多实时可用数据的潜力。基于此，本论文针对混合交通流提出一个由网联车协助的数据融合框架以检测状态参数，训练基于BP（Back Propagation，反向传播）神经网络的路侧传感器数据融合模型以应用于包括CVs和RVs的所有车辆。考虑到路况实时变化，本文提出一种动态BP融合方法来对交通流状态准确检测。同时，由于车辆日益增多，数据量逐渐增大，导致边缘服务器MEC（Mobile Edge Computing）工作负载过大。基于此，本论文提出一种优先级反向计算卸载决策，以最小化系统平均消耗。建立了基于优先级和场景坐标的马尔可夫决策过程（Markov Decision Process，MDP）模型，以确保时延敏感任务高效完成。本文的主要研究内容如下：

（1）调研多种路侧传感器，确定以毫米波雷达、摄像头数据作为融合模型输入，重点研究基于熵的贝叶斯方法和传统BP融合，通过仿真验证融合结果。分析了混合交通流中智能网联车所发挥的重要作用。搭建了网联车协助的路侧传感器（毫米波雷达和智能摄像头）数据融合框架以适用于混合交通流状态检测。该框架主要利用 CVs 实时V2X（Vehicle to Everything）通信获得的相对更准确的交通参数作为标准值来训练BP神经网络，将基于BP的数据融合模型应用于包括RVs在内的所有车辆。

（2）此外，考虑复杂的道路环境实时变化，构建了一种动态BP融合模型以进一步实现在长时间内保持参数融合结果的准确性。分别对交通流量、占有率、速度参数融合模型进行仿真，最后计算误差评价指标和准确率指标。仿真结果验证了三种融合方法的误差均有所降低，与单传感器检测、基于熵的贝叶斯融合方法和传统的BP融合相比，所提出的动态BP融合方法更准确，且三类交通参数的融合结果准确率均达到95%以上。

（3）在多个网联车辆和单个移动边缘计算服务器MEC的单向车道场景中，提出了基于优先级的反向计算卸载模式，每一辆车都需要将自身的感知数据在一定时隙内上传至RSU（Road Side Unit，路侧单元），由RSU决策任务是否需要进行反向卸载，以减少MEC工作负担。采用深度强化学习算法研究反向卸载决策优化，以降低系统平均消耗为目标，确保按时完成高优先级车辆的时延敏感性高的任务。并将该决策问题根据车辆优先级和场景坐标系建立了MDP模型。最后，对比了四种方案在不同仿真参数（包括MEC的CPU周期频率、权重因子、计算量、决策周期及优先级参数）下的结果，验证了基于优先级的反向卸载方案具有最优性能。

Due to the emergence of vehicles with different levels of intelligence, more and more connected vehicles (CVs) and regular vehicles (RVs) will be mixed on the road, which is the necessary stage of future traffic development. Accurate traffic flow state is particularly important to ensure safe and efficient traffic. At present, multi-sensors have been deployed on a large scale to sense the road traffic environment. In addition, the ability for connected vehicles on the road to communicate with others could potentially yield more real-time and reliable data. Based on this, this paper proposes a data fusion framework, aided by intelligent connected vehicles, to detect state parameters in mixed traffic flow. It trains the roadside sensor data fusion model, based on BP (Back Propagation) neural network, to be applicable to all vehicles, including CVs and RVs. Considering the real-time changes of road conditions, a dynamic BP fusion method is proposed to accurately detect the traffic flow state. At the same time, due to the increasing number of vehicles and the increasing amount of data, the workload of the edge server MEC (Mobile Edge Computing) is too large. Based on this, a priority reverse computing offloading decision is proposed to minimize average system consumption. A Markov Decision Process (MDP) model based on priority and scene coordinates is established to ensure the efficient completion of delay-sensitive tasks. The following are the main study contents of the paper:

(1) Investigate a variety of roadside sensors, determine to adopt millimeter wave radar and camera data as the input of the fusion model. The focus is on the entropy-based Bayesian method and the traditional BP fusion, and the fusion results are verified by simulation. The important role of intelligent CVs in mixed traffic flow is analyzed. A data fusion framework of roadside sensors (millimeter wave radar and intelligent camera) assisted by the connected vehicle has been built to apply to the detection of mixed traffic flow. This framework mainly adopts the relatively more accurate traffic parameters obtained by CVs real-time V2X (Vehicle to Everything) communication as standard values to train BP neural network, and applies the data fusion model based on BP to all vehicles including RVs.

(2) In addition, considering the real-time changes of complex road environment, a dynamic BP fusion model is constructed to further maintain the accuracy of parameter fusion results for a long time. Simulate the fusion model of traffic flow, occupancy, and speed parameters respectively, and calculate error evaluation indicators and accuracy indicators at the end. The simulation results confirm that the errors of the three fusion methods have been decreased. Compared with single-sensor detection, entropy-based Bayesian fusion method and traditional BP fusion, the proposed dynamic BP fusion method is more accurate, and the accuracy of the fusion results of the three types of traffic parameters is more than 95%.

(3) In the one-way lane scenario of multiple connected vehicles and a single mobile edge computing server MEC, a priority based reverse computing offloading mode is proposed. Each vehicle needs to upload its own perception data to the roadside RSU (Road Side Unit) within a certain time slot. The RSU decides whether the task needs to be offloaded in reverse to reduce the workload of MEC. Adopting the deep reinforcement learning algorithm, the aim is to optimize reverse offloading decisions, thus reducing the system's average consumption and guaranteeing that tasks with high delay sensitivity of high-priority vehicles are completed on time. The MDP model is established according to the vehicle priority and scene coordinate system. Finally, the results of four schemes under different simulation parameters (including the CPU cycle frequency of MEC, weight factor, calculation amount, decision cycle and priority parameter) are compared, verifying that the reverse offloading scheme based on priority has the optimal performance.

[1]Qureshi K N, Abdullah A H. A survey on intelligent transportation systems[J]. Middle-East Journal of Scientific Research, 2013, 15(5): 629-642.
[2]Shaaban K, Elamin M, Alsoub M. Intelligent transportation systems in a developing country: Benefits and challenges of implementation[J]. Transportation Research Procedia, 2021, 55: 1373-1380.
[3]Xie D F, Zhao X M, He Z. Heterogeneous traffic mixing regular and connected vehicles: Modeling and stabilization[J]. IEEE Transactions on Intelligent Transportation Systems, 2018, 20(6): 2060-2071.
[4]Yang J S. Travel time prediction using the GPS test vehicle and Kalman filtering techniques[C]//Proceedings of the 2005, American Control Conference, 2005. IEEE, 2005: 2128-2133.
[5]El Faouzi N E. Data fusion in road traffic engineering: An overview[J]. Multisensor, Multisource Information Fusion: Architectures, Algorithms, and Applications 2004, 2004, 5434: 360-371.
[6]Golestan K, Sattar F, Karray F, et al. Localization in vehicular ad hoc networks using data fusion and V2V communication[J]. Computer Communications, 2015, 71: 61-72.
[7]Chen S, Hu J, Shi Y, et al. A vision of C-V2X: Technologies, field testing, and challenges with chinese development[J]. IEEE Internet of Things Journal, 2020, 7(5): 3872-3881.
[8]胡明伟, 张芷铭, 陈湘生. 智能网联车辆混行交通流效益研究[J]. 系统仿真学报, 2021, 33(9): 2270.
[9]Zhou H, Wu T, Chen X, et al. Reverse auction-based computation offloading and resource allocation in mobile cloud-edge computing[J]. IEEE Transactions on Mobile Computing, 2022.
[10]Sun F, Hou F, Cheng N, et al. Cooperative task scheduling for computation offloading in vehicular cloud[J]. IEEE Transactions on Vehicular Technology, 2018, 67(11): 11049-11061.
[11]Habiba U, Maghsudi S, Hossain E. A reverse auction model for efficient resource allocation in mobile edge computation offloading[C]//2019 IEEE global communications conference (GLOBECOM). IEEE, 2019: 1-6.
[12]Liwang M, Dai S, Gao Z, et al. A truthful reverse-auction mechanism for computation offloading in cloud-enabled vehicular network[J]. IEEE Internet of Things Journal, 2018, 6(3): 4214-4227.
[13]He H, Jiang H. Deep learning based energy efficiency optimization for distributed cooperative spectrum sensing[J]. IEEE Wireless Communications, 2019, 26(3): 32-39.
[14]Wiering M A, Van Otterlo M. Reinforcement learning[J]. Adaptation, learning, and optimization, 2012, 12(3): 729.
[15]Arvin R, Khattak A J, Kamrani M, et al. Safety evaluation of connected and automated vehicles in mixed traffic with conventional vehicles at intersections[J]. Journal of Intelligent Transportation Systems, 2020, 25(2): 170-187.
[16]Chang X, Li H, Rong J, et al. Analysis on traffic stability and capacity for mixed traffic flow with platoons of intelligent connected vehicles[J]. Physica A: Statistical Mechanics and Its Applications, 2020, 557: 124829.
[17]Matcha B N, Sivanesan S, Ng K C. Modeling lane-changing behavior of vehicles at merge section under mixed traffic conditions[J]. Journal of transportation engineering, Part A: Systems, 2021, 147(4): 04021006.
[18]胡满江, 卜令坤, 秦洪懋等. 多类时延下混合车辆队列建模与协同控制[J]. 汽车工程, 2022(009):044.
[19]Rizvi S M A, Ahmed A, Shen Y. Real-time incident detection and capacity estimation using loop detector data[J]. Journal of Advanced Transportation, 2020, 2020: 1-14.
[20]Zhao J, Xu H, Liu H, et al. Detection and tracking of pedestrians and vehicles using roadside LiDAR sensors[J]. Transportation research part C: emerging technologies, 2019, 100: 68-87.
[21]Tan Q J O, Romero R A. Ground vehicle target signature identification with cognitive automotive radar using 24–25 and 76–77 GHz bands[J]. IET Radar, Sonar & Navigation, 2018, 12(12): 1448-1465.
[22]Kerner B S, Demir C, Herrtwich R G, et al. Traffic state detection with floating car data in road networks[C]//Proceedings. 2005 IEEE Intelligent Transportation Systems, 2005. IEEE, 2005: 44-49.
[23]Li X, She Y, Luo D, et al. A traffic state detection tool for freeway video surveillance system[J]. Procedia-Social and Behavioral Sciences, 2013, 96: 2453-2461.
[24]Ahmadi S A, Ghorbanian A, Mohammadzadeh A. Moving vehicle detection, tracking and traffic parameter estimation from a satellite video: A perspective on a smarter city[J]. International journal of remote sensing, 2019, 40(22): 8379-8394.
[25]Lee G H, Choi J D, Lee J H, et al. Object detection using vision and LiDAR sensor fusion for multi-channel V2X system[C]//2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC). IEEE, 2020: 1-5.
[26]Treiber M, Kesting A, Wilson R E. Reconstructing the traffic state by fusion of heterogeneous data[J]. Computer‐Aided Civil and Infrastructure Engineering, 2011, 26(6): 408-419.
[27]Anand R A, Vanajakshi L, Subramanian S C. Traffic density estimation under heterogeneous traffic conditions using data fusion[C]//2011 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2011: 31-36.
[28]Tang X, Zhang Z, Qin Y. On-road object detection and tracking based on radar and vision fusion: A review[J]. IEEE Intelligent Transportation Systems Magazine, 2022, 14(5).
[29]Jibrin F A, Deng Z, Zhang Y. An object detection and classification method using radar and camera data fusion[C]//2019 IEEE International Conference on Signal, Information and Data Processing (ICSIDP). IEEE, 2019: 1-6.
[30]Yu Z, Bai J, Chen S, et al. Camera-radar data fusion for target detection via Kalman filter and Bayesian estimation[R]. SAE Technical Paper, 2018.
[31]Ren J, Wang Y, Han Y, et al. Information fusion of digital camera and radar[C]//2019 IEEE MTT-S International Microwave Biomedical Conference (IMBioC). IEEE, 2019, 1: 1-4.
[32]Kumar R, Jayashankar S. Radar and camera sensor fusion with ROS for autonomous driving[C]//2019 Fifth International Conference on Image Information Processing (ICIIP). IEEE, 2019: 568-573.
[33]Park S, Jang J A, Kim J. Infrastructure based vehicle recognition system with multi sensor fusion[C]//2013, 7th International Conference on Signal Processing and Communication Systems (ICSPCS). IEEE, 2013: 1-5.
[34]Storck C R, Duarte-Figueiredo F. A survey of 5G technology evolution, standards, and infrastructure associated with vehicle-to-everything communications by internet of vehicles[J]. IEEE access, 2020, 8: 117593-117614.
[35]Chen S, Hu J, Shi Y, et al. Vehicle-to-everything (V2X) services supported by LTE-based systems and 5G[J]. IEEE Communications Standards Magazine, 2017, 1(2): 70-76.
[36]Garcia M H C, Molina-Galan A, Boban M, et al. A tutorial on 5G NR V2X communications[J]. IEEE Communications Surveys & Tutorials, 2021, 23(3): 1972-2026.
[37]Chen S, Hu J, Shi Y, et al. A vision of C-V2X: Technologies, field testing, and challenges with chinese development[J]. IEEE Internet of Things Journal, 2020, 7(5): 3872-3881.
[38]Barrachina J, Sanguesa J A, Fogue M, et al. V2X-d: A vehicular density estimation system that combines V2V and V2I communications[C]//2013 IFIP Wireless Days (WD). IEEE, 2013: 1-6.
[39]Zhao X, Mu K, Hui F, et al. A cooperative vehicle-infrastructure based urban driving environment perception method using a DS theory-based credibility map[J]. Optik, 2017, 138: 407-415.
[40]Chen X, Jiao L, Li W, et al. Efficient multi-user computation offloading for mobile-edge cloud computing[J]. IEEE/ACM transactions on networking, 2015, 24(5): 2795-2808.
[41]Xiao Z, Dai X, Jiang H, et al. Vehicular task offloading via heat-aware MEC cooperation using game-theoretic method[J]. IEEE Internet of Things Journal, 2019, 7(3): 2038-2052.
[42]Dolezal J, Becvar Z, Zeman T. Performance evaluation of computation offloading from mobile device to the edge of mobile network[C]//2016 IEEE Conference on Standards for Communications and Networking (CSCN). IEEE, 2016: 1-7.
[43]Hu H, Wang Q, Hu R Q, et al. Mobility-aware offloading and resource allocation in a MEC-enabled IoT network with energy harvesting[J]. IEEE Internet of Things Journal, 2021, 8(24): 17541-17556.
[44]Zhou H, Jiang K, Liu X, et al. Deep reinforcement learning for energy-efficient computation offloading in mobile-edge computing[J]. IEEE Internet of Things Journal, 2021, 9(2): 1517-1530.
[45]Li J, Gao H, Lv T, et al. Deep reinforcement learning based computation offloading and resource allocation for MEC[C]//2018 IEEE wireless communications and networking conference (WCNC). IEEE, 2018: 1-6.
[46]Jia Z, Zhou Z, Wang X, et al. Learning-based queuing delay-aware task offloading in collaborative vehicular networks[C]//ICC 2021-IEEE International Conference on Communications. IEEE, 2021: 1-6.
[47]Alfakih T, Hassan M M, Gumaei A, et al. Task offloading and resource allocation for mobile edge computing by deep reinforcement learning based on SARSA[J]. IEEE Access, 2020, 8: 54074-54084.
[48]Chen J, Xing H, Xiao Z, et al. A DRL agent for jointly optimizing computation offloading and resource allocation in MEC[J]. IEEE Internet of Things Journal, 2021, 8(24): 17508-17524.
[49]Shi J, Du J, Wang J, et al. Distributed V2V computation offloading based on dynamic pricing using deep reinforcement learning[C]//2020 IEEE Wireless Communications and Networking Conference (WCNC). IEEE, 2020: 1-6.
[50]Yang H, Wei Z, Feng Z, et al. Intelligent computation offloading for MEC-based cooperative vehicle infrastructure system: A deep reinforcement learning approach[J]. IEEE Transactions on Vehicular Technology, 2022, 71(7): 7665-7679.
[51]Feng W, Yang S, Gao Y, et al. Reverse offloading for latency minimization in vehicular edge computing[C]//ICC 2021-IEEE International Conference on Communications. IEEE, 2021: 1-6.
[52]Feng W, Zhang N, Li S, et al. Latency minimization of reverse offloading in vehicular edge computing[J]. IEEE Transactions on Vehicular Technology, 2022, 71(5): 5343-5357.
[53]Gu A, Wu H, Tang H, et al. Deep reinforcement learning-guided task reverse offloading in vehicular edge computing[C]//GLOBECOM 2022-2022 IEEE Global Communications Conference. IEEE, 2022: 2200-2205.
[54]Sun Z, Guo M, Liu W, et al. Multisource traffic data fusion with entropy based method[C]//2009 International Conference on Artificial Intelligence and Computational Intelligence. IEEE, 2009, 4: 506-509.
[55]Pan X, Zhou W, Lu Y, et al. Prediction of network traffic of smart cities based on DE-BP neural network[J]. IEEE Access, 2019, 7: 55807-55816.
[56]Gudise V G, Venayagamoorthy G K. Comparison of particle swarm optimization and backpropagation as training algorithms for neural networks[C]//Proceedings of the 2003 IEEE Swarm Intelligence Symposium. SIS'03 (Cat. No. 03EX706). IEEE, 2003: 110-117.
[57]Li Y. Deep reinforcement learning: An overview[J]. arXiv preprint arXiv:1701.07274, 2017.
[58]LeCun Y, Bengio Y, Hinton G. Deep learning[J]. nature, 2015, 521(7553): 436-444.
[59]Dayan P, Niv Y. Reinforcement learning: The good, the bad and the ugly[J]. Current opinion in neurobiology, 2008, 18(2): 185-196.
[60]杨思明, 单征, 丁煜等. 深度强化学习研究综述[J]. 计算机工程, 2021, 47(12): 19-29.
[61]Wang X, Wang S, Liang X, et al. Deep reinforcement learning: A survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022.
[62]Clifton J, Laber E. Q-learning: Theory and applications[J]. Annual Review of Statistics and Its Application, 2020, 7: 279-301.
[63]Mnih V, Kavukcuoglu K, Silver D, et al. Playing atari with deep reinforcement learning[J]. arXiv preprint arXiv:1312.5602, 2013.
[64]Mo Y, Zhang P, Chen Z, et al. A method of vehicle-infrastructure cooperative perception based vehicle state information fusion using improved Kalman filter[J]. Multimedia Tools and Applications, 2022: 1-18.
[65]Qin A K, Huang V L, Suganthan P N. Differential evolution algorithm with strategy adaptation for global numerical optimization[J]. IEEE transactions on Evolutionary Computation, 2008, 13(2): 398-417.