Multilane Energy-efficient Predictive Cruise Control For Commercial Vehicle Using Value-function Learning

Our country’s economic prosperity promotes the vigorous development of the transportation industry.The vehicle ownership has increased rapidly,which has caused the traffic congestion and a large amount of energy consumption.The rapid development of the computer technology promotes the progress of network and algorithm.Real-time transmission of the information from two vehicles and the evolution of autonomous vehicle algorithms has become possible.This paper designs a multilane energy-efficient predictive cruise control system with the background of automatic driving and aims to improve the efficiency under the premise of safety.In this paper,firstly,I develop an algorithm for solving optimal lane changing and acceleration based on the model predictive control.Secondly,the acceleration quadratic optimization algorithm and output torque solving algorithm based on value-function learning are developed.Finally,the validity,reliability and universality of the algorithm are verified by simulation and hardware implementation.The development and verification of multilane energy-efficient predictive cruise control system has been completed.Compared with the traditional CC and ACC,this algorithm increases the freedom of lateral lane change.When there is a vehicle with low speed in front of you,you can change lanes to maintain the original driving state to improve the economy under the premise of safety.The main research contents of this paper are as follows:(1)The design and modeling of the energy-efficient predictive cruise control system considering the lane change decisions.Firstly,the overall architecture design of the energy-efficient predictive cruise control system is completed.Secondly,the force of the commercial vehicle is analyzed and the longitudinal dynamics model is established.Thirdly,the energy consumption model is established and the equation is obtained by analyzing the relationship between the output power,the output speed and the output torque.Fourthly,a three-lane model and a two-lane model are established to provide model support for subsequent optimization.Finally,the vehicle kinematics model is established to calculate and predict the future state of each vehicle in the prediction time domain.(2)Development and verification of lane change and acceleration optimization algorithm based on model predictive control.Firstly,the corresponding objective function of mixed integer programming is established based on multi-objective programming.Secondly,the corresponding constraints are proposed based on the actual vehicle speed limit,acceleration limit and safety limit.Thirdly,a simple three-lane model is established to verify the effectiveness of the algorithm,and it is found that the vehicle can choose the correct lane change time and stable acceleration under the premise of safety.Fourthly,a complex two-lane model is established to verify the reliability of the algorithm,and it is found that the vehicle can accelerate and change lanes normally within the set time.Finally,a complex three-lane model is established to verify the universality of the algorithm,and it is found that vehicles can correctly select the optimal lane and slow down to ensure safety when there is a slow vehicle blocking and no lane for it to change.Therefore,the proposed algorithm is effective,reliable and universal.(3)Development and verification of acceleration quadratic optimization and torque optimization algorithm based on value-function learning.Based on the optimal acceleration output from the above algorithm,the optimal control sequence of torque can be obtained by value-function learning algorithm.Firstly,the corresponding objective function of mixed integer programming is established based on multi-objective programming.Secondly,the corresponding constraints are proposed based on the actual vehicle speed limit,acceleration limit and safety limit.A speed tracking scenario is established to verify the effectiveness of the algorithm.Compared with the results of dynamic programming,the accuracy of the value-function learning algorithm meets the requirements.An acceleration scenario is established to verify the reliability and universality of the algorithm.Compared with the results of dynamic programming,the accuracy of the value-function learning algorithm meets the requirements.The algorithm can be optimized correctly at the same initial or the terminal speed.Finally,the validity of the algorithm is verified by hardware based on vehicle tracking conditions.