| Global navigation satellite system is often used as a means of obtaining vehicle locations because of its advantages such as convenience,maturity and low cost,but it cannot overcome the influence of vehicle mobility,receiver noise and multipath interference.In order to solve the problem of inaccurate vehicle positioning in the global navigation satellite system,this paper relies on the national key research and development program of China "Theory and Test Verification of Vehicle Group Intelligent Control in Vehicle-Road Collaborative Environment"(2018YFB1600600),studies the vehicle positioning and roadside unit deployment algorithm based on V2 X from the perspective of roadside unit assisted positioning:(1)Aiming at the problem of declining vehicle positioning accuracy of global navigation satellite systems in urban neighborhoods,a V2 X vehicle positioning method based on extended Kalman filter is designed.The real-time motion information sent by the target vehicle to the roadside unit and the relative distance of the vehicle perceived by the roadside unit itself are fused by an filter to improve the accuracy of the roadside unit in estimating the position of the moving vehicle;the update of the filter is triggered by the received information packets or the solver’s solution coordinates to avoid the impact of packet loss on the robustness of the system in a short period of time;considering that the filter prediction and update process the dimensionality of the correlation matrix will reach an upper limit,a replacement metric is defined to measure the optimal set of tracked vehicles.The results show that the method achieves high positioning accuracy in both urban and suburban scenarios,outperforming the fusion of global navigation satellite systems position information with inertial navigation system measurement data.(2)Aiming at the actual situation where the sensing range and computing power of roadside unit is limited,and in order to provide assisted positioning services to as many vehicles as possible while ensuring the working efficiency of roadside unit,a parameter sharing and updating mechanism for the pre-training error prediction model is proposed to achieve the correction of all vehicle positioning errors in the vicinity of roadside unit.Given that global navigation satellite system error is the result of joint interference from multiple factors,a stacking ensemble learning framework is used to build an error prediction model;support vector regression,extreme gradient boosting,and Catboost are selected as base learners;and lasso regression is implemented as a meta-learner to obtain further generalisation results;compared with random forest and adaptive boosting,the proposed method achieves the best performance.Taking the error prediction of the east-west axis of Zhangjiapu scene as an example,the average absolute error decreases by 11.34% and 10.79%,the root mean square error decreases by 31.3%and 24.26%,and the average absolute percentage error decreases by 24.08% and 38.03%,respectively.(3)In order to use the roadside unit assisted positioning services in the complex urban road network,a roadside unit deployment method based on genetic algorithm and cuckoo local optimization is designed.A numerical sequence is chosen to encode the roadside unit candidate locations for the genetic algorithm;a discrete cuckoo search is applied to the child individuals created by the genetic algorithm;the proposed algorithm is validated based on real city road map layouts and vehicle trajectories.The results show that the average fitness score and the best fitness score of the proposed method are significantly higher than those of existing genetic algorithms and memetic algorithm,and provide a basis for the deployment of different numbers of roadside units. |
