Department of Power and Control Engineering, School of Electrical and Computer Engineering, Shiraz University, Shiraz 71348-51154, Iran.
Department of Power and Control Engineering, School of Electrical and Computer Engineering, Shiraz University, Shiraz 71348-51154, Iran.
ISA Trans. 2018 Mar;74:134-143. doi: 10.1016/j.isatra.2018.02.005. Epub 2018 Feb 16.
This paper proposes two novel Kalman-based learning algorithms for an online Takagi-Sugeno (TS) fuzzy model identification. The proposed approaches are designed based on the unscented Kalman filter (UKF) and the concept of dual estimation. Contrary to the extended Kalman filter (EKF) which utilizes derivatives of nonlinear functions, the UKF employs the unscented transformation. Consequently, non-differentiable membership functions can be considered in the structure of the TS models. This makes the proposed algorithms to be applicable for the online parameter calculation of wider classes of TS models compared to the recently published papers concerning the same issue. Furthermore, because of the great capability of the UKF in handling severe nonlinear dynamics, the proposed approaches can effectively approximate the nonlinear systems. Finally, numerical and practical examples are provided to show the advantages of the proposed approaches. Simulation results reveal the effectiveness of the proposed methods and performance improvement based on the root mean square (RMS) of the estimation error compared to the existing results.
本文提出了两种基于卡尔曼滤波的新的在线 Takagi-Sugeno（TS）模糊模型辨识学习算法。所提出的方法基于无迹卡尔曼滤波器（UKF）和对偶估计的概念设计。与利用非线性函数导数的扩展卡尔曼滤波器（EKF）不同，UKF 采用了无迹变换。因此，可以在 TS 模型的结构中考虑不可微的隶属函数。这使得所提出的算法能够应用于更广泛类别的 TS 模型的在线参数计算，而与最近关于同一问题的发表的论文相比。此外，由于 UKF 在处理严重非线性动力学方面的强大能力，所提出的方法可以有效地逼近非线性系统。最后，提供了数值和实际示例来说明所提出方法的优点。仿真结果显示了所提出方法的有效性，并与现有结果相比，基于估计误差的均方根（RMS），提高了性能。
