Construction And Field Testing Of Personalized Car-following Guidance System For Connected Vehicles

As the source and transmission path of traffic oscillations,car-following behavior is a key factor that affects the stability of traffic flow.Due to the limitations in perception,control,and decision-making of human drivers,human car-following behavior is often unstable and easily induces traffic oscillations.The emerging vehicle-to-everything(V2X)communication provides an opportunity to improve car-following behavior.Accurate perception of vehicle operating status and information sharing beyond visual range can effectively improve the rationality of car-following behavior and alleviate traffic oscillations.Based on this,cooperative adaptive cruise control and queue control technologies have been deeply studied in intelligent connected vehicle cooperative control technology and have demonstrated strong capabilities in preventing the spread of traffic oscillations and improving traffic efficiency.However,the manufacturing and retrofitting costs of intelligent connected vehicles are relatively high and cannot be popularized in the short term.For connected vehicles(CVs)with low cost and easy to popularize compared to human-driven vehicles,existing advanced driver assistance systems(ADAS)mainly focus on safety aspects such as collision and curve speed warnings.Further research is needed on how to use the capabilities of CVs to improve the car-following efficiency of human drivers.Based on the research needs mentioned above,this article constructs a vehicle-to-vehicle(V2V)follow-up guidance system based on connected vehicles.Through on-site testing with real vehicles,this study verifies some follow-up guidance methods based on intelligent connected vehicle control algorithms.At the same time,considering the personalized characteristics of different drivers,this study further constructs a personalized follow-up guidance system based on LSTM(long short-term memory)using measured data and conducts actual tests.The specific work of this article is as follows:(1)In response to the need for building a follow-up guidance system,this article first developed a primary connected vehicle(CV)follow-up guidance system based on hardware devices such as global navigation satellite system(GNSS)facilities,dedicated short-range communication(DSRC)devices,and personal computers.The system integrates the above hardware and is designed in a modular way at the software level.The system is divided into three modules: data acquisition,data processing,and human-machine interface(HMI).The data acquisition module obtains GNSS data through the serial port,marks and sends the data through UDP communication,and the data processing module is responsible for inducing speed generation and data storage.All of these functions are implemented in C++.In addition,this study developed an HMI for publishing guidance information based on web development.(2)Based on the verification needs of the primary car-following guidance system,this study conducted real-vehicle tests at the Changan University Intelligent Vehicle and Transportation System Testbed.At the algorithm level,this primary car-following guidance system includes four longitudinal guidance algorithms based on car-following models and intelligent networked vehicle control.At the HMI level,this study designed a comparative experiment between systems with and without guidance information.Fifty volunteers were recruited for the study,and their driving behaviors were collected and analyzed based on the given scenarios.The results showed that the nondiscriminatory guidance information provided by the primary networked vehicle car-following guidance system can significantly improve the driver’s ability to follow and maintain a stable distance from the preceding vehicle.(3)The article proposes a new personalized following induction algorithm based on Long Short-Term Memory(LSTM)neural network to meet the personalized induction needs of drivers.The algorithm couples human factors with the Cooperative Adaptive Cruise Control(CACC)controller of intelligent connected vehicles to achieve personalized response to different drivers.To integrate the existing C++-based following induction system with the Python-based personalized induction algorithm across languages and platforms,this study uses a database as a relay node and develops a new HMI based on the Flask framework for publishing induction information.(4)Regarding the analysis of following behavior under personalized guidance,this experiment selected 10 volunteers and continued to conduct on-road tests at the Changan University’s Connected Vehicle and Intelligent Vehicle Test Field.The driving behavior of the participants was collected and analyzed based on given scenarios.This study designed a comparative experiment of "consistent","grouping",and "personalized" guidance models based on different training set ranges,as well as an iterative experiment that included time dimension training data iteration.The experimental results showed that the guidance model with a smaller training set range had a better effect on improving following efficiency,and after one iteration,the "personalized" guidance model performed better in improving driver following behavior.In summary,this thesis constructed a primary follow-up induction information system and a personalized induction system,designed experimental schemes for on-road tests,and studied and analyzed the impact of induction information on driving efficiency in the networked vehicle environment.The software and hardware system built and the personalized follow-up induction model constructed can provide prototypes,technical references,and theoretical support for future networked vehicle applications.