基于深度强化学习的车辆多目标协同巡航决策控制系统设计
2023,31(10):115-121
摘要：为提升车辆巡航避障能力，实现对运动目标的精准决策，设计基于深度强化学习的车辆多目标协同巡航决策控制系统。利用主控制电路输出的电量信号，调节ACC控制器、MPC轨迹跟踪器、双闭环控制器的实时连接状态，再借助多目标解耦模块，确定目标车辆所处巡航位置，完成巡航决策控制系统的主要应用结构设计。建立深度强化学习模型，根据车辆目标数据集定义条件，求解协同参数实际取值范围，实现对车辆巡航位姿的估计。确定坐标转换原则，通过分析多目标量化结果的方式，实现对巡航决策轨迹的按需规划，再联合相关应用设备，完成基于深度强化学习的车辆多目标协同巡航决策控制系统的设计。实验结果表明，深度强化学习机制作用下，车辆在横、纵两个巡航方向上的避障准确度都达到了100%，符合车辆多目标协同巡航决策的实际需求。
关键词：深度强化学习；车辆多目标；协同巡航；决策控制；轨迹跟踪器；多目标解耦；协同参数；量化分析；
Design of vehicle multi-objective cooperative cruise decision control system based on deep reinforcement learning
Abstract：In order to improve the vehicle cruise obstacle avoidance ability and achieve accurate decision-making on moving targets, a vehicle multi-target collaborative cruise decision-making control system based on deep reinforcement learning is designed. The electricity signal output from the main control circuit is used to adjust the real-time connection state of ACC controller, MPC track tracker and double closed loop controller. Then, with the help of multi-objective decoupling module, the cruise position of the target vehicle is determined, and the main application structure design of the cruise decision-making control system is completed. The depth reinforcement learning model is established. According to the definition conditions of vehicle target data set, the actual value range of collaboration parameters is solved to realize the estimation of vehicle cruise pose. Determine the coordinate conversion principle, realize the on-demand planning of the cruise decision-making trajectory by analyzing the multi-target quantitative results, and then combine relevant application equipment to complete the design of the vehicle multi-target collaborative cruise decision-making control system based on in-depth reinforcement learning. The experimental results show that under the deep reinforcement learning mechanism, the obstacle avoidance accuracy of the vehicle in both horizontal and vertical cruising directions reaches 100%, which meets the actual requirements of vehicle multi-target cooperative cruise decision-making.
Key words：intensive learning; Vehicle multi-target; Cooperative cruise; Decision control; Track tracker; Multi objective decoupling; Collaboration parameters; Quantitative analysis;
收稿日期：2022-12-16
基金项目：2022年度广西中青年教师科研基础能力提升项目：基于强化学习的智能车决策算法研究，2022KY0606
