作　　者： ; ;

机构地区： 东北大学机械工程与自动化学院

出　　处： 《华中科技大学学报（自然科学版）》 2015年第1期7-11,共5页

摘　　要： 为了有效地控制液压非线性系统,提出基于RBF神经网络的自适应最优控制系统,应用于机器人液压驱动器.首先,建立了液压系统的动力学模型;然后,输入幅值和频率连续变化的信号,应用卡尔曼滤波器估计液压系统状态,进而计算出模型参数,对模型参数进行分组用于训练RBF神经网络;接着,对不同组参数求平均作为参考点,用RBF神经网络学习最优控制器反馈增益随系统参数的变化规律;最后,训练完成的神经网络根据卡尔曼滤波器参数估计值在线预测并调节控制器增益.经实验验证,该控制系统调节时间和跟踪误差仅为普通线性二次型最优控制器的1/2和1/3左右. In order to effectively control the hydraulic nonlinear systems,a radial basis function（RBF）neural network-based optimal control applied to the robot hydraulic actuator was presented.First,the hydraulic servo system was modeled based on the physics of the plant.Second,the Kalman filter was applied to estimate the internal state of the system with continuously changing magnitude and frequency of input signal.The model parameters were calculated and grouped for RBF neural network training.Third,with the average of each group of parameters as nominal point,the RBF neural network was used to learn the rules how feedback gains changes with system parameters.Finally,the trained neural network was used to predict the feedback gains on line based on the parameter estimate of Kalman Filter and the trained adaptive controller.The proposed controller was validated by experiment with setting time and tracking error to be 1/2and 1/3of the conventional linear quadratic requlator controller,respectively.

关 键 词： 卡尔曼滤波器 自适应控制 神经网络 最优控制 液压驱动器

领　　域： [自动化与计算机技术] [自动化与计算机技术]