论文标题
LPV模型的调度尺寸缩小 - 深度神经网络方法
Scheduling Dimension Reduction of LPV Models -- A Deep Neural Network Approach
论文作者
论文摘要
在本文中，审查了线性参数变化（LPV）模型的现有调度尺寸缩小（SDR）方法，并开发了深层神经网络（DNN）方法，该方法在降低调度维度下实现了更高的模型准确性。在模型的准确性和与减少模型合成的控制器的模型准确性和性能方面，在两链机器人操作器上比较了所提出的DNN方法和现有的SDR方法。比较的方法包括使用主组件分析（PCA），内核PCA（KPCA）和自动编码器（AE）的状态空间模型的SDR。在机器人操纵器的示例中，与当前方法相比，DNN方法以Frobenius Norm的形式获得了原始LPV模型的基质变化的改进表示。此外，当将结果模型用于适应合成时，与当前方法相比，获得了改进的闭环性能。
In this paper, the existing Scheduling Dimension Reduction (SDR) methods for Linear Parameter-Varying (LPV) models are reviewed and a Deep Neural Network (DNN) approach is developed that achieves higher model accuracy under scheduling dimension reduction. The proposed DNN method and existing SDR methods are compared on a two-link robotic manipulator, both in terms of model accuracy and performance of controllers synthesized with the reduced models. The methods compared include SDR for state-space models using Principal Component Analysis (PCA), Kernel PCA (KPCA) and Autoencoders (AE). On the robotic manipulator example, the DNN method achieves improved representation of the matrix variations of the original LPV model in terms of the Frobenius norm compared to the current methods. Moreover, when the resulting model is used to accommodate synthesis, improved closed-loop performance is obtained compared to the current methods.
