基于深度哈密顿神经网络的物料提升机的鲁棒控制

Robust control of elevator based on Hamiltonian neural network

安装基座的随机振动给物料提升机系统带来不确定性,对此提出一种基于深度神经网络的非线性鲁棒控制策略。首先,采用一种嵌入哈密顿力学先验的深度神经网络(HNN),拟合了系统的动力学模型;然后,以此模型为基础,结合一种隐式Lyapunov(IL)函数,设计了系统的鲁棒镇定控制器。仿真显示,所提出的基于深度哈密顿神经网络结合IL函数控制器(HIL)的控制效果与基于精确模型的控制器IL几乎完全一致;相较传统的基于黑箱神经网络模型的IL控制(BIL)、基于误差模型(IL-E),其收敛时间分别减少了25.9%和32.5%。结果表明,所提出的控制策略能准确表征系统的非线性动力学特征,有效抑制基座振动不确定的影响,实现物料提升机的快速精确鲁棒镇定。

A nonlinear robust control strategy based on deep neural network was proposed for the uncertain elevator with random oscillation at the mounting base.First,a deep neural network(HNN)with Hamiltonian mechanics priors was employed to fit the dynamic model of the system.Then,based on this model,a robust stabilization controller for the system was designed by combining an implicit Lyapunov(IL)function.Simulation results showed that the proposed controller(HIL)achieved almost identical control performance to the controller(IL)based on the accurate model.Compared with traditional IL control based on black-box neural network models and error models,the convergence time of the proposed controller was reduced by 25.9%(BIL)and 32.5%(IL-E),respectively.The results demonstrated that the proposed control strategy could accurately represent the nonlinear dynamic characteristics of the system,effectively suppress the uncertain influence of base oscillation,and achieve rapid and accurate robust stabilization of the elevator.