基于多模型自适应方法的智能汽车路径跟踪控制

路径跟踪控制是智能汽车的一项核心技术，跟踪效果的精确性和在各种路面附着条件下的鲁棒性是该技术的两大关键要素。但汽车动力学模型的不确定性，尤其是轮胎侧偏刚度的摄动使这两者难以同时得到满足。针对这一问题，将多模型自适应理论引入到智能汽车运动控制中处理不确定性系统的控制。首先，推导了多模型自适应控制律，提出了凸包构架下各个顶点的子模型对真实模型的自适应逼近律，并通过李雅普诺夫函数证明了所提出自适应律的收敛能力。在此基础上建立了汽车动力学模型和车辆-路径联合模型，并由多个顶点子模型构建可覆盖汽车轮胎侧偏刚度摄动范围的凸多面体，利用汽车动力学模型求解自适应率，通过车辆-路径联合模型，基于线性二次型方法(linear quadratic regulator, LQR)求解各个顶点的子模型处的反馈控制律，并通过所得出的自适应权重进行加权。基于Carsim/Simulink的联合仿真结果表明，所提出的多模型自适应路径跟踪控制器在保证鲁棒性的同时克服了传统鲁棒控制方法的保守性问题，与基于名义模型的LQR控制器和鲁棒保性能控制器相比，在高附着路面和低附着路面上都可以取得更好的控制效果，很好地解决了路径跟踪...

Path following control of intelligent vehicles based on multi-model adaptive method

Path following control is a crucial technology for intelligent vehicles, and the control accuracy and the robustness under various road adhesive conditions are two key elements of this technology. However, the accuracy and the robustness are hard to be achieved simultaneously owing to the uncertainties in a vehicle dynamics model, especially the perturbation of tire cornering stiffness. To deal with the uncertainties, a multi-model adaptive method is introduced in this study. Firstly, the basic theory of the method is derived, and the adaptive law of each vertex sub-model to the real model is proposed, with its convergence proved by the Lyapunov theory. Then, a vehicle dynamics model and a vehicle-road combined model are built, and the convex polyhedron including all possible perturbation of tire cornering stiffness is established with multiple sub-models. The adaptive law is derived according to the vehicle dynamics model, and the feedback controller of the sub-model in each vertex is derived by the linear quadratic regulator (LQR) method based on the vehicle-road combined model. Simulation results show that the proposed controller can not only ensure the robustness, but also overcome the conservative problem of previous robust methods, achieving excellent performance under various road conditions. Finally, a rapid prototyping test platform is established for further evaluation. Results show that the proposed algorithm has excellent real-time performance, suggesting an excellent potential of its engineering application.