基于数据驱动的高速列车速度复合控制研究

Data-driven Speed Compound Control of High-speed Train

针对高速列车在外部干扰下的速度控制问题,本文提出基于Koopman算子的高速列车高维线性模型的建模方法,并设计一种结合扩张状态观测器(ESO)与基于Koopman算子的模型预测控制(K-MPC)的复合控制器(ESO-K-MPC)。利用扩展动态模式分解算法来近似无限维线性Koopman算子,建立具有动态非线性特性的高速列车动力学高维线性模型;引入模型预测控制,设计扩张状态观测器,对系统总扰动进行估计与补偿,构建基于ESO-K-MPC的高速列车速度控制系统,再设计控制器与控制算法;结合CRH3列车参数和郑西高铁华山北站—西安北站实际线路数据,分别在没有扰动和白噪声干扰下对设计的控制方法与算法进行仿真研究。仿真结果表明:基于Koopman的高速列车建模对位移与速度的预测精度相比于线性状态空间模型分别提高了83.86%与87.40%;ESO-K-MPC可以准确估计与补偿高速列车运行中受到的干扰,控制输出曲线与期望曲线几乎重叠,实现了列车运行期望曲线的高精度跟踪。

Targeting the topic of high-speed train speed control amid external disturbances,a modeling method of a high-dimensional linear model of high-speed trains based on the Koopman operator is proposed,and a composite controller(ESO-K-MPC)combining extended state observer(ESO)and model predictive control(K-MPC)based on Koopman operator is designed.Firstly,the dynamic high-dimensional linear model of a high-speed train with dynamic nonlinear characteristics was developed by approximating the infinite-dimensional linear Koopman operator using the extended dynamic mode decomposition algorithm.Secondly,the model predictive control was implemented,and the extended state observer was built to estimate and correct the system's total disturbance.The ESO-K-MPC-based highspeed train speed control system was created,together with the controller and control algorithm.Finally,combined CRH3 train parameters with the actual line data of Huashan North Station,Xi'an North Station of Zhengzhou-Xi'an High-speed Railway,the designed control methods and algorithms were simulated and studied without disturbance and white noise interference respectively.The simulation indicates that the displacement and velocity forecast accuracy of high-speed train modeling based on Koopman is increased by 83.86 and 87.40%,respectively,when compared to the linear state space model.ESO-K-MPC can accurately estimate and compensate interference during high-speed train operation,the control output curve practically overlaps the expected curve,and high precision tracking of the predicted curve during train operation is achieved.