基于多智能体理论的汽车底盘协调控制方法研究

发布时间：2018-10-15 19:42

【摘要】：随着人类社会科技文明的高度发展,生活节奏的不断加快,汽车作为代步工具的时代逐步到来。为了满足乘坐舒适、操纵方便、安全可靠和人类不断追求完美的要求,汽车产品己由初期的完全机械结构发展到现在融合机械、电子、材料、控制等多学科和科技新成果应用的阶段,而且功能越来越强,可靠性越来越高,正朝着多目标综合和智能化控制的方向发展。其中,与车辆乘坐舒适性、操纵稳定性和行驶安全性能密切相关的悬架、制动和转向等底盘协调控制系统成为汽车工程领域研究的热点。本文依据汽车的动力学原理,在MATLAB/Simulink环境下,结合多传感器采集得到的相关数据,建立汽车底盘各智能体数学模型,包括制动Agent模型、转向Agent模型和7自由度悬架Agent模型。然后,将RBF神经网络的强化学习算法用于底盘多智能体的协调控制。通过小脑神经模型(CMAC)算法先对悬架Agent、转向Agent和制动Agent提供的大量连续的数据进行状态量的约简泛化,作为强化学习算法的状态输入。根据底盘各智能体的性能指标确定强化信号,并利用RBF神经网络训练强化学习算法的动作网络和评价网络,实现本地智能体的性能指标最优。其次,采用模糊控制方法确定强化学习中性能指标权重的大小。根据不同工况下的数据制定经验规则来确定各智能体性能指标的权重,完成汽车底盘多智能体强化学习控制策略的设计,提高汽车的综合性能。最后,在MATLAB/Simulink仿真环境下,搭建以xPC Target实时的仿真系统硬件在环试验台,利用所设计的多智能体强化学习控制算法,针对汽车的制动行驶工况、瞬间转向行驶工况以及转向制动复杂行驶工况进行仿真实验,并对结果进行详细的分析,验证了本文所提出的多智能体协调控制算法在提高汽车舒适性、安全性以及平顺性方面的有效性。
[Abstract]:With the high development of science and technology civilization of human society and the quickening of life rhythm, the era of automobile as a walking tool is coming gradually. In order to meet the requirements of comfortable ride, convenient operation, safety and reliability and the continuous pursuit of perfection by human beings, automobile products have developed from the initial complete mechanical structure to the present integration of machinery, electronics and materials. The application stage of multi-disciplinary and new achievements in science and technology, such as control, is developing towards the direction of multi-objective synthesis and intelligent control, and the function is more and more powerful and the reliability is getting higher and higher. Among them, suspension, braking, steering and other chassis coordination control systems, which are closely related to vehicle ride comfort, handling stability and driving safety, have become a hot spot in the field of automotive engineering. In this paper, according to the dynamic principle of automobile, the mathematical models of automobile chassis agents, including braking Agent model, steering Agent model and 7-degree-of-freedom suspension Agent model, are established under the environment of MATLAB/Simulink and the relevant data collected by multi-sensor. Then, the reinforcement learning algorithm of RBF neural network is applied to the coordinated control of chassis multi-agent. The cerebellar neural model (CMAC) algorithm is used to reduce and generalize a large number of continuous data provided by suspension Agent, steering Agent and brake Agent as the state input of reinforcement learning algorithm. According to the performance index of each agent in chassis, the enhancement signal is determined, and the action network and evaluation network of reinforcement learning algorithm are trained by RBF neural network to realize the best performance index of local agent. Secondly, fuzzy control method is used to determine the weight of performance index in reinforcement learning. According to the data under different working conditions, the weight of each agent performance index is determined according to the data under different working conditions, and the design of the reinforcement learning control strategy of multi-agent chassis is completed, and the comprehensive performance of the vehicle is improved. Finally, under the MATLAB/Simulink simulation environment, the hardware of the xPC Target real-time simulation system is built on the ring test rig, and the designed multi-agent reinforcement learning control algorithm is used to solve the braking condition of the vehicle. The simulation experiments on instantaneous steering conditions and complex steering braking conditions are carried out, and the results are analyzed in detail. It is verified that the multi-agent coordinated control algorithm proposed in this paper can improve the vehicle comfort. Effectiveness in terms of safety and ride comfort.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.1;TP18

【参考文献】