基于最优结构多维泰勒网的含噪声非线性时变系统辨识
发布时间：2018-12-30 21:28
【摘要】：针对具有噪声干扰的非线性时变系统建模时存在的困难,建立了一种具有最优结构和最佳泛化能力的多维泰勒网模型,以实现对该系统的辨识.首先,为了能够快速反映系统输入/输出的变化,以多维泰勒网的连接权系数作为时变参数,并由带可变遗忘因子的递推最小二乘算法对其进行训练,进而讨论了辨识方案的稳定性.然后,为了避免维数灾难并满足实时性要求,给出了选择多维泰勒网有效回归项的改进权衰减法,以使多维泰勒网同时具有最小结构和最佳的泛化能力.最后,通过算例说明基于最优结构的多维泰勒网在含噪声非线性时变系统辨识问题中应用的方法,算例结果验证了该方法的有效性.
[Abstract]:Aiming at the difficulties in modeling nonlinear time-varying systems with noise disturbance, a multi-dimensional Taylor net model with optimal structure and optimal generalization ability is established to identify the system. Firstly, in order to reflect the change of input / output of the system quickly, the connection weight coefficient of multi-dimension Taylor net is taken as the time-varying parameter, and it is trained by the recursive least squares algorithm with variable forgetting factor. Furthermore, the stability of the identification scheme is discussed. Then, in order to avoid dimensionality disaster and meet the real-time requirements, an improved weight attenuation method for selecting effective regression terms of multi-dimensional Taylor nets is presented, so that the multi-dimensional Taylor nets have the minimum structure and the best generalization ability at the same time. Finally, an example is given to illustrate the application of multi-dimensional Taylor nets based on optimal structure to the identification of nonlinear time-varying systems with noise. The effectiveness of the proposed method is verified by an example.
【作者单位】： 东南大学自动化学院;河南工学院计算机科学与技术系;东南大学复杂工程系统测量与控制教育部重点实验室;
【基金】：国家自然科学基金资助项目(61673112,60934008) 中央高校基本科研业务费专项资金资助项目(2242017K10003,2242014K10031) 江苏高校优势学科建设工程资助项目
【分类号】：TB53
本文编号：2396173
[Abstract]:Aiming at the difficulties in modeling nonlinear time-varying systems with noise disturbance, a multi-dimensional Taylor net model with optimal structure and optimal generalization ability is established to identify the system. Firstly, in order to reflect the change of input / output of the system quickly, the connection weight coefficient of multi-dimension Taylor net is taken as the time-varying parameter, and it is trained by the recursive least squares algorithm with variable forgetting factor. Furthermore, the stability of the identification scheme is discussed. Then, in order to avoid dimensionality disaster and meet the real-time requirements, an improved weight attenuation method for selecting effective regression terms of multi-dimensional Taylor nets is presented, so that the multi-dimensional Taylor nets have the minimum structure and the best generalization ability at the same time. Finally, an example is given to illustrate the application of multi-dimensional Taylor nets based on optimal structure to the identification of nonlinear time-varying systems with noise. The effectiveness of the proposed method is verified by an example.
【作者单位】： 东南大学自动化学院;河南工学院计算机科学与技术系;东南大学复杂工程系统测量与控制教育部重点实验室;
【基金】：国家自然科学基金资助项目(61673112,60934008) 中央高校基本科研业务费专项资金资助项目(2242017K10003,2242014K10031) 江苏高校优势学科建设工程资助项目
【分类号】：TB53
【相似文献】
相关期刊论文 前1条
1 彭海波;于开平;刘炜;;基于改进最小二乘步骤的NARMA模型辨识非线性时变结构系统[J];噪声与振动控制;2010年02期
，本文编号：2396173
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2396173.html
