柔性机翼长基线天线变形测量的模糊网络法研究

发布时间：2018-05-18 08:11

本文选题：模糊网络 + 机翼变形测量　；参考：《西安电子科技大学》2015年硕士论文

【摘要】：高空长航时无人机的发展日益受到重视,而对其机翼长基线天线变形的实时高精度测量对保证天线性能具有重要意义。本文将模糊理论应用到机翼变形测量的实际问题中,基于Takagi-Sugeno-Kang(TSK)型模糊逻辑系统的普遍逼近特性,提出两种不基于被测对象的新型模糊网络算法,使其能够更为准确有效地逼近应变量和形变位移量二者之间的关系,具有较强的推理能力、自适应学习能力和通用性,为机翼变形测量提供一种新的思路。本文首先针对自构架模糊网络法(Self-Structuring Fuzzy Network,SSFN)抗干扰性较差的缺点,结合神经网络和区间二型模糊理论,提出一种新型自构架区间二型模糊神经网络(Self-Structuring Interval Type-2 Fuzzy Neural Network,SSIT2FNN)。在SSIT2FNN中,规则前件为区间二型模糊集合,用于将每条规则的激活强度反馈到自身构成内反馈回路,其参数学习采用梯度下降算法;后件为带有区间权值的TSK模型,其参数学习采用有序规则卡尔曼滤波算法。网络的初始规则数为零,并且所有规则均通过结构学习和前后件参数同时在线学习来产生。由于SSIT2FNN是基于训练误差最小化的思路来进行网络的训练学习,在网络的构建过程中往往依赖于训练经验,从而使得系统泛化能力较差。因此,本文又结合聚类分裂的思想和支持向量回归(Support Vector Regression,SVR)理论,提出一种自分裂迭代线性支持向量回归模糊网络(Self-Splitting Iterative Linear SVR Fuzzy Network,SSILSVRFN)。在SSILSVRFN中,网络的构建过程主要分为结构学习和参数学习两个部分,其中,结构学习采用自分裂规则生成算法(Self-Splitting Rule Generation,SSRG)来自动生成规则并对其初始化;参数学习则基于结构风险最小化原则,采用迭代线性SVR(Iterative Linear SVR,ILSVR)的学习算法来对规则的前、后件参数进行迭代优化。并针对同一非线性函数进行逼近仿真,深入对比分析了不同方法之间的特点。最后,自主设计机翼框架模型并搭建测量实验平台,完成机翼框架模型在不同静态载荷下的变形测量实验。实验结果表明,本文提出的两种改进型模糊网络算法实现了对逼近精度的逐步提高,能够较为准确的反映出应变量和形变位移量二者之间的关系,并初步验证了模糊网络变形测量方法在实际中的有效性。
[Abstract]:The development of UAV has been paid more and more attention during long altitude navigation, and it is very important to measure the deformation of wing long baseline antenna with high precision in real time to ensure the antenna performance. In this paper, the fuzzy theory is applied to the practical problem of wing deformation measurement. Based on the general approximation property of Takagi-Sugeno-Kangn TSK-based fuzzy logic system, two new fuzzy network algorithms are proposed, which are not based on the measured object. It can more accurately and effectively approach the relationship between strain and deformation displacement. It has strong reasoning ability, adaptive learning ability and generality, which provides a new way for wing deformation measurement. In this paper, first of all, aiming at the disadvantage of the self-Structuring Fuzzy Network (SSFN) method, combining the neural network and the interval type 2 fuzzy theory, a new self-structuring Interval Type-2 Fuzzy Neural network SSIT2FNN is proposed. In SSIT2FNN, the former part of the rule is interval type 2 fuzzy set, which is used to feedback the activation intensity of each rule to the inner feedback loop. The parameter learning adopts gradient descent algorithm, and the latter part is the TSK model with interval weight. The order rule Kalman filter algorithm is used for parameter learning. The number of initial rules of the network is zero, and all the rules are generated by learning the structure and the parameters simultaneously online. Because SSIT2FNN is based on the idea of minimizing the training error, it often relies on the training experience in the process of network construction, which makes the generalization ability of the system poor. Therefore, combining the idea of clustering splitting with the support vector regression support Vector regress (SVR) theory, a self-splitting iterative linear support vector regression fuzzy network called Self-Splitting Iterative Linear SVR Fuzzy Network is proposed in this paper. In SSILSVRFN, the construction process of the network is mainly divided into two parts: structure learning and parameter learning. The self-splitting Rule generation algorithm (SSRG) is used to automatically generate and initialize the rules. Parameter learning is based on the principle of structural risk minimization, and iterative linear SVR(Iterative Linear SVR (ILSVR) learning algorithm is used to optimize the parameters of the first and last parts of the rules. The approximation simulation of the same nonlinear function is carried out, and the characteristics of different methods are compared and analyzed in depth. Finally, the wing frame model is designed and the experimental platform is built to measure the deformation of the wing frame model under different static loads. The experimental results show that the proposed two improved fuzzy network algorithms can improve the approximation accuracy step by step, and can accurately reflect the relationship between strain and deformation displacement. The effectiveness of the fuzzy network deformation measurement method in practice is preliminarily verified.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V279

【参考文献】