Cooperative Ramp Merging Control Considering The Uncertainty Of Human Vehicles

Connected automated vehicle(CAV)is one of the important development directions of intelligent transportation system in the future.However,it will take at least 40-50 years for connected automated vehicles to mix with traditional human driving vehicles and low automated vehicles to achieve a high penetration or even 100% CAV environment.Therefore,this paper mainly studies the cooperative ramp merging decision-making of connected automated vehicles in mixed traffic flow under the uncertainty of traditional human driving vehicles.This paper solves this problem by using model predictive control system and human driving behavior prediction based on deep learning,which correspond to the second,third and fourth chapters respectively.Firstly,in the second chapter of this paper,a closed-loop optimal control method based on model predictive control is proposed.The control method enhances the robustness of the algorithm to the uncertain behavior of human driving vehicles by real-time sampling,optimization and feedback of vehicle trajectory data.The case of ramp merging is used to illustrate the application of the proposed hybrid centralized-decentralized model predictive control framework.In this framework,the Road Side Unit(RSU)will be used as a centralized infrastructure,which is mainly responsible for the trajectory prediction of human driven vehicles and the decision-making of vehicle ramp merging sequence considering the global optimization.The distributed on-board computing unit(OBU)is mainly responsible for its own optimal cooperative ramp merging trajectory planning.Moreover,this paper discusses in detail how the centralized equipment and distributed equipment in the framework interactively transmit decision-making instructions and vehicle information.Secondly,the ramp merging sequence problem and the cooperative ramp merging trajectory planning problem involved in the framework can be described as two optimization problems.This paper uses the method of bilevel programming to solve them.The upper-level merge sequencing problem is solved using a dynamic-programming-based approach.Considering real-time computational performance,three solution methods,including dynamic programming(DP),dynamic matrix predictive control(DMC),and simplified discrete control,are developed to solve the lower-level trajectory planning problem.Thirdly,a human driving vehicle trajectory prediction model based on Graph Attention Neural Network(GAT)is constructed.The model uses the Long Short Term Memory Network(LSTM)encoder-decoder model,a graph attention neural network module,in which the LSTM is used to process the characteristics of time series,and the GAT module is used to accurately learn the spatial interaction between vehicles.By using the graph neural network based on attention aggregator,the developed model can learn the attention paid by drivers to the vehicles in their local neighborhood and aggregate forward traffic flow information,thereby significantly improving the long-term prediction accuracy and the interpretability of the model.Next Generation Simulation(NGSIM)datasets are used to train and evaluate our model.The results indicate that we reduce the root mean square error(RMSE)by 17%according to quantitative evaluation.Finally,this paper uses SUMO traffic flow simulation software,and redevelops the longitudinal car following model to increase the randomness of car following as the experimental environment.The results show that compared with open-loop control,the proposed hybrid model predictive control(HMPC)framework can effectively address the uncertainty caused by the stochastic driving behaviors of HVs.In addition,it is found that hybrid centralized-decentralized control can reduce the computational time to the millisecondlevel and provide a similar,system-efficient performance compared to centralized control,potentially meeting real-time computational requirements.