基于在线高斯模型驱动MPC的四旋翼轨迹跟踪控制  

Trajectory tracking control for a quadrotor based on online Gaussian model-driven MPC

作　　者：叶大鹏[1] 陈书达 张之得 YE Dapeng;CHEN Shuda;ZHANG Zhide(College of Mechanical and Electrical Engineering,Fujian Agriculture and Forestry University,Fuzhou 350100,China)

机构地区：[1]福建农林大学机电工程学院,福建福州350100

出　　处：《飞行力学》2025年第1期56-62,共7页Flight Dynamics

基　　金：福建省中青年教师教育科研项目资助(JAT210082)。

摘　　要：针对四旋翼飞行器轨迹跟踪控制中模型预测控制(MPC)的标称模型不确定问题,提出了一种基于在线高斯过程回归模型增强的模型预测控制(OGP-MPC)方法,利用在线高斯过程回归(OGP)模型补偿标称模型的动力学误差。设计了一种新的在线GP模型更新框架,通过引入子GP模型对新数据进行预处理,提高数据质量,进而迭代更新主GP模型参数,以实现自适应动力学模型误差补偿。仿真结果表明,相比传统MPC和GP-MPC,所提方法在圆形轨迹下的模型精度和跟踪精度提升均超过16%,空间曲线轨迹下提升超过5%。An online Gaussian process regression driven model predictive control(OGP-MPC)ap-proach is proposed to address the issue of uncertainty in nominal dynamic models of model predictive control(MPC)for quadrotor trajectory tracking.The method utilizes online Gaussian process regres-sion(OGP)models to compensate for the dynamic errors of the nominal models.A novel online GP model updating framework is designed,which involves preprocessing new data with sub-GP models to improve data quality and iteratively updating the parameters of the main GP model to achieve a-daptive compensation for dynamic model errors.Simulation results reveal that compared to traditional MPC and GP-MPC,the proposed method improves model accuracy and tracking precision by over 16%under circular trajectories,and more than 5%under spatial curved trajectories.

关 键 词：四旋翼 模型预测控制 数据驱动 高斯过程回归 轨迹跟踪 

分 类 号：V279[航空宇航科学与技术—飞行器设计] V249.1