基于小波与对角神经网络的陀螺误差建模及其应用研究

发布时间：2018-05-21 07:55

本文选题：组合导航 + 惯性元件　；参考：《哈尔滨工程大学》2014年硕士论文

【摘要】：在工程领域,无论何种形式的对象,信息感知都是一项重要的课题。信息的感知可以是感知自己,感知对方或者感知环境。导航就是一种信息感知。无论何种载体,若想要到达目的地,都需要有导航信息来辅助运动控制。所以,导航的精度至关重要。为提升导航系统的精度和可靠性,人们提出了组合导航系统,通过组合不同种类的子导航系统,以克服子系统的局限并实现精度的提升。在组合导航系统中,惯导系统往往处于主导地位,而惯导的精度又由惯性元件决定。与此同时,为进行组合导航的数据融合,需要根据各子系统构建精确的数学模型,否则将会使组合导航变得毫无优势可言。因此,惯性元件误差的精确模型,对于误差的补偿或者组合导航系统的构建均具有重要意义。针对这一点,本文从组合建模的观点出发,组合使用小波阈值去噪和对角神经网络对惯性元件随机误差进行建模。并在基于伪距、伪距率的组合导航系统中使用了这种模型以证明其有效性。首先,对INS/GNSS组合导航相关的模型进行了介绍和推导,其中包括惯导误差方程、GNSS相关原理以及组合导航系统的构建等。并且在模型推导时,尽量不做简化处理,保留更完整的参数信息。其次,使用Allan方差和功率谱密度对光纤陀螺的实测数据进行分析,并发现使用传统的随机数、相关噪声加白噪声的假设模型不足以精确描述其随机误差,而若使用ARMA模型则操作复杂且表现力不足。为此,根据分析所得,本文选用小波阈值去噪对陀螺进行一步消噪处理。在小波阈值去噪时,文中提供了一个求取分解层次参考值的方法并对比了各种阈值去噪规则的效果。仿真结果表明,在消噪后,陀螺的中、高频噪声被有效滤除。再次,针对去噪后残留的陀螺随机误差具有低频和相关性的特点,使用对角神经网络进行时序建模。为加速网络的收敛,使用LM算法改进了对角神经网络的学习算法;讨论了对角神经网络时序建模的优势,证明了使用对角神经网络更方便快捷。在接受模型检验后,对角神经网络的模型残差为零均值的白噪声,这也将为组合导航滤波带来了便利。仿真结果表明,使用对角神经网络对陀螺这些残留随机误差建模取得了良好的效果。最后,根据文中所使用的方法和结论,提出一种小波阈值去噪和对角神经网络建模辅助下的组合导航系统,并使用跑车实测数据进行仿真验证。结果表明,在这种模型辅助下,基于伪距、伪距率的组合导航系统有更好的表现。
[Abstract]:In the field of engineering, no matter what form of object, information perception is an important subject. The perception of information can be to perceive oneself, to perceive the other party, or to perceive the environment. Navigation is a kind of information perception. No matter what kind of carrier, to reach the destination, we need navigation information to assist motion control. Therefore, the accuracy of navigation is very important. In order to improve the accuracy and reliability of the navigation system, the integrated navigation system is proposed, which can overcome the limitations of the subsystem and improve the accuracy by integrating different kinds of sub-navigation systems. In integrated navigation systems, inertial navigation systems are often dominant, and the accuracy of inertial navigation is determined by inertial components. At the same time, in order to fuse the data of integrated navigation, it is necessary to construct accurate mathematical models according to each subsystem, otherwise, the integrated navigation will have no advantage at all. Therefore, the accurate model of inertial element error is of great significance for error compensation or integrated navigation system construction. In this paper, from the point of view of combinatorial modeling, wavelet threshold denoising and diagonal neural network are used to model the random error of inertial elements. This model is used in the integrated navigation system based on pseudo-range and pseudo-range rate to prove its validity. Firstly, the related models of INS/GNSS integrated navigation are introduced and deduced, including the ins error equation and the related principle of GNSS, and the construction of integrated navigation system. And in the derivation of the model, as far as possible do not do simplification processing, retain more complete parameter information. Secondly, the Allan variance and power spectral density are used to analyze the measured data of fog, and it is found that the assumption model of correlation noise and white noise is not enough to describe the random error accurately. If the ARMA model is used, the operation will be complicated and the performance will be insufficient. Therefore, according to the analysis results, the wavelet threshold de-noising is used to remove the gyroscope. In the wavelet threshold denoising, a method to obtain the reference value of decomposition hierarchy is provided, and the effects of various threshold denoising rules are compared. The simulation results show that the high frequency noise of the gyroscope is effectively filtered after de-noising. Thirdly, aiming at the low frequency and correlation of the residual random error after denoising, the diagonal neural network is used to model the time series. In order to accelerate the convergence of the network, the learning algorithm of diagonal neural network is improved by using LM algorithm, and the advantage of time series modeling of diagonal neural network is discussed, which proves that using diagonal neural network is more convenient and fast. After the model is tested, the model residual of the diagonal neural network is white noise with zero mean value, which will also facilitate the integrated navigation filtering. The simulation results show that the diagonal neural network is used to model these residual random errors of gyroscope. Finally, according to the methods and conclusions used in this paper, a new integrated navigation system based on wavelet threshold denoising and diagonal neural network modeling is proposed. The results show that the integrated navigation system based on pseudo-range and pseudo-range rate has better performance under the aid of this model.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN96;TP183

【参考文献】