机器人关节电机的广义预测控制

Generalized Predictive Control for Robot Joint Motors

机器人关节电机控制系统具有非线性和参数变化的特点,基于被控对象精准数学模型的传统控制方法难以对其进行有效的控制。以四足机器人髋关节电机为研究对象,首先分析了系统的机理,建立了被控对象的CARIMA(Controlled Auto-Regressive Integrated Moving Average,受控自回归积分滑动平均)模型;接着提出了一种基于复合神经网络的广义预测控制方法,即由LNN(Linear Neural Network,线性神经网络)和GPFN(Gaussian Poten-tial Function Networks,高斯基函数网络)构成的复合网络对被控对象进行在线辨识,广义预测控制器利用辨识的结果,多步预测,滚动优化,对四足机器人髋关节电机的角位移进行有效控制;最后假定系统存在Stribeck型非线性摩擦,在负载转动惯量缓慢变化和突变的情况下进行了仿真试验,结果表明,该方法具有较强的适应能力,体现了很强的鲁棒性,取得了令人满意的控制效果。

As the characteristics of the nonlinearity and parameter variety exist in the robot joint motors, the traditional control methods based on the precise model of the controlled plant can not be able to acl effectively. In this paper,aiming at the sciatic joint motor of the quadruped robot, firstly the systemic mechanism was analyzed, and the CARIMA（ controlled auto-regressive integrated moving average） model of the plant was established. Secondly a novel generalized predictive control （GPC） method based on composite neural network was brought forward,indicating that the composite neural network composed of a LNN （linear neural network） and a GPFN（ gaussian potential function networks） was used for on-line system identification. The generalized predictive controller, making use of the result from the composite neural network identifier, predicted for several steps, roll-optimized and controlled the angle displacement of the joint motor. Finally supposing the Stribeck friction existed in the plant,the method was simulated with the parameter-the loading moment of inertia varying slowly and quickly. The results indicate that the proposed method acquires a good performance with stronger adaptive capacity and stronger robustness.