【作者】 王杰；

【导师】 付尧；

【作者基本信息】 吉林大学 ， 机械（专业学位）， 2024， 硕士

【摘要】 重型商用车作为道路交通运输支柱对于国民经济发展具有重要作用,其能源消耗大、碳排放占比高,燃油消耗成本占商用车全生命周期(Total Cost of Ownership,TCO)的1/3以上,因而实现重型商用车的节能降耗具有重要意义。汽车智能网联化发展为整车节能控制提供了新契机,预见性自适应巡航控制(Predictive Adaptive Cruise Control,PACC)是重型商用车节能控制的重要手段,其协同自车速度、前方道路坡度、限速、前车运动状态以显著降低整车能耗。车速规划与挡位优化是PACC的关键技术,考虑执行层动力传动系统的节能特性及PACC场景切换与功能分层下的变时间尺度规划控制问题仍有待深入,基于此本文开展了基于改进深度强化学习算法的节能车速规划、基于非线性模型预测控制的速度跟踪控制及基于非线性规划的挡位优化控制的研究。具体研究内容如下:(1)分析了重型商用车动力传动系统结构,建立了重型商用车纵向动力学模型,制定了其经济性换挡规律;然后针对速度规划控制求解获得的最优驱动力建立了挡位优化控制器,将发动机万有特性曲线拟合为解析表达,用于求解节能的变速器挡位,最后对发动机转速转矩与挡位进行合理性检查,当结果不合理时按照经济性换挡规律执行换挡,并将控制器应用至纵向动力学模型进行验证。(2)针对预见性自适应巡航考虑前方道路与车辆状态信息的问题,对双延迟深度确定性策略梯度算法进行剖析并提出结合预测信息的改进算法。根据PACC有前车的自适应巡航跟驰与无前车的预见性巡航对预测时域需求的差异,分为两个场景分别进行节能的速度规划。在自适应巡航跟驰场景,由于前车行为的不确定性,采取短时域并考虑安全性的节能车速规划。在预见性巡航场景,由于没有前车的干扰,因此主要考虑前方道路坡度对车辆的节能影响,在更长的时间域内实现节能车速规划。(3)由于规划与控制要求的时间尺度不同,将规划与控制分层,使用非线性模型预测控制完成考虑坡度的速度跟踪控制任务。面向场景不同的控制需求,在自适应巡航跟驰场景中更多地考虑安全与舒适的限制,对前车的车速进行基于马尔可夫链的预测,提出基于深度强化学习的参考速度轨迹生成方法,并设置满足最大安全制动距离的安全车距以及车速限制,以保证乘员的安全与舒适体验。在预见性巡航场景中,由于没有前车影响,更多地考虑速度跟踪特性,兼顾安全与舒适,实现更好的节能效果。(4)基于Python与Py Torch建立深度强化学习算法模型,对Highway-Env环境进行纵向动力学集成,并使用Cas ADi作为求解器求解非线性规划与非线性模型预测控制问题,实现联合仿真训练,并设置典型工况对算法进行了仿真与验证。仿真结果表明,所提出的算法对重型商用车PACC燃油经济性有一定提升。更多还原

【Abstract】 Heavy-duty commercial vehicles,as the backbone of road transportation,are considered to play a crucial role in the development of the national economy.However,they exhibit high energy consumption and carbon emissions,with fuel costs constituting over one-third of the Total Cost of Ownership(TCO)for commercial vehicles throughout their lifecycle.Therefore,achieving energy savings and consumption reduction in heavy-duty commercial vehicles is of significant importance.A new opportunity for enhancing vehicle energy efficiency control is presented by the development of intelligent and connected vehicle systems in the automotive industry.Predictive Adaptive Cruise Control(PACC)emerges as a pivotal approach for energy control in heavy-duty commercial vehicles,as it effectively coordinates vehicle speed with road gradients,speed limits,and the movement status of preceding vehicles,resulting in notable reductions in overall vehicle energy consumption.Key techniques within PACC include speed planning and gear optimization.However,challenges persist in addressing the energy-saving characteristics of the powertrain system and the temporal scale planning control problem under PACC scenario switching and functional layering.In this context,this paper conducts research on energy-saving speed planning based on improved deep reinforcement learning algorithms,speed tracking control based on nonlinear model predictive control,and gear optimization control based on nonlinear programming.The specific research content is as follows:(1)Analysis is conducted on the structure of the powertrain system in heavy-duty commercial vehicles,establishing a longitudinal dynamics model,and formulating the economic shifting law.Subsequently,an optimization controller for the powertrain system is developed based on the optimal driving force obtained from speed planning control solutions,which is used to determine energy-efficient engine operating points and gear positions.Rationality checks are performed on engine speed,torque,and gear positions.When the results are deemed irrational,shifting is executed according to the economic shifting law.The optimization controller is then applied to the longitudinal dynamics model for validation.Finally,a basic description of the speed planning control problem in the predictive time domain,an energy estimation model suitable for optimization solutions,and objective functions are provided.(2)An analysis of the dual-delay deep deterministic policy gradient algorithm is conducted,considering the anticipatory adaptive cruise control system that incorporates information on road and vehicle states,and an improved algorithm combining predictive information is proposed.Based on the differences in predictive time domain requirements between adaptive cruise control with preceding vehicles and predictive cruise control without preceding vehicles,energy-saving speed planning is divided into two scenarios.In the adaptive cruise control scenario,due to the uncertainty of preceding vehicle behavior,short-term domain energy-saving speed planning considering safety is adopted.In the predictive cruise control scenario,where there is no preceding vehicle interference,the focus is primarily on the energy-saving impact of road gradients,achieving energy-saving speed planning over a longer time horizon.(3)A hierarchical approach is employed due to the different time scales required for planning and control,using nonlinear model predictive control to accomplish speed tracking control considering gradients.Catering to varying control requirements across different scenarios,more emphasis is placed on safety and comfort constraints in the adaptive cruise control scenario.Predictive models based on Markov chains are used to forecast preceding vehicle speeds,and a reference speed trajectory generation method based on deep reinforcement learning is proposed.Safety distances satisfying maximum braking distance and speed restrictions are adopted to ensure passenger safety and comfort.In the predictive cruise control scenario,with no preceding vehicle influence,speed tracking characteristics are prioritized to achieve better energy-saving effects while maintaining safety and comfort.(4)Python and Py Torch are used to establish deep reinforcement learning algorithm models.Highway-Env environment integration for longitudinal dynamics is conducted,and Cas ADi is utilized as a solver for solving nonlinear programming and nonlinear model predictive control problems.Joint simulation training is implemented,and the proposed algorithms are simulated and validated under typical scenarios and operating conditions.Simulation results demonstrate a certain improvement in fuel economy for heavy-duty commercial vehicles with PACC.更多还原