加速度传感器参数非线性时间序列模型预测与实现

发布时间：2018-06-02 14:35

  本文选题：加速度表参数 + NAR神经网络　； 参考：《西南科技大学》2016年硕士论文

【摘要】：加速度表广泛应用于惯性导航系统中,加速度表精度的提高对提高惯性导航控制精度具有很高的重要性。为提高加速度表的精度,一方面可以进一步改进加速度表的设计和生产工艺,另一方面可以对加速度表的参数进行补偿。经研究发现,加速度表参数的时间序列具有很强的非线性特点,使用传统的线性模型,难以达到很高的建模精度。因此本论文提出对加速度表参数进行非线性时间序列建模,以提高建模和预测精度。本论文以石英挠性加速度表为研究对象,研究并设计了加速度表静态参数的标定方法,编写了加速度表数据采集软件系统,以获得研究所需要的加速度表的零偏和标度因数;由于加速度表参数的时间序列具有很强的非线性特点,提出了使用非线性时间序列进行建模。在非线性时间序列算法研究中,在传统BP神经网络的基础上增加延时环节使其具有记忆历史数据的能力,建立了NAR神经网络对参数进行建模;在AR模型的基础上加入小波神经网络,建立了小波神经网络与AR组合模型,该模型利用AR模型拟合其线性部分,用小波神经网络拟合其非线性部分,并对小波神经网络的算法进行改进,使得该模型具有收敛速度快,训练效果好。由于各种模型的预测方法都有其独特的信息特征和适用条件,本论文利用组合预测理论,吸收每种模型的优点,提出使用改进型贝叶斯组合预测方法,对NAR神经网络和小波神经网络与AR组合模型的预测结果进行组合预测。利用该方法与传统ARMA模型的预测结果进行对比表明组合预测效果较好,准确率较高。
[Abstract]:Speedup meter is widely used in inertial navigation system. It is very important to improve the accuracy of inertial navigation system by improving the precision of accelerometer. In order to improve the accuracy of the accelerometer, the design and production process of the accelerometer can be further improved on the one hand, and the parameters of the accelerometer can be compensated on the other hand. It is found that the time series of accelerometer parameters have strong nonlinear characteristics and it is difficult to achieve high modeling accuracy by using the traditional linear model. In order to improve the accuracy of modeling and prediction, nonlinear time series modeling of accelerometer parameters is proposed in this paper. This paper takes the quartz flexible accelerometer as the research object, studies and designs the calibration method of the static parameters of the accelerometer, compiles the data acquisition software system of the accelerometer, in order to obtain the zero bias and scale factor of the accelerometer needed by the research. Because the time series of accelerometer parameters have strong nonlinear characteristics, a nonlinear time series is proposed to model the model. In the research of nonlinear time series algorithm, the NAR neural network is established to model the parameters by adding the delay link to the traditional BP neural network so that it has the ability to memorize the historical data. The combination model of wavelet neural network and AR is established by adding wavelet neural network on the basis of AR model. The model uses AR model to fit its linear part, and wavelet neural network to fit its nonlinear part. The algorithm of wavelet neural network is improved, which makes the model convergent quickly and has good training effect. Because the forecasting methods of various models have their unique information characteristics and applicable conditions, this paper uses the combination forecasting theory to absorb the advantages of each model, and proposes an improved Bayesian combination forecasting method. The prediction results of NAR neural network and wavelet neural network combined with AR are predicted. The comparison between this method and the traditional ARMA model shows that the combined prediction is effective and accurate.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP212
，

本文编号：1969113