一种基于多尺度和改进支持向量机的光纤陀螺温度漂移建模与补偿方法

为了提升光纤陀螺温度漂移模型建模的准确性及补偿的效果,提出了一种基于改进支持向量机的多尺度建模和回归方法。首先分析了造成光纤陀螺温度漂移的关键因素,给出了建模的属性参数和温度试验。然后根据经验模态分解得到的本征模态函数排列熵的变化趋势,得出了回归精度和熵之间的变化关系,进而提出了基于信号分解的多尺度回归方法。为了提高上述多尺度回归算法的适应性,在传统支持向量机的基础上,提出了基于组合核函数的支持向量机回归算法,以适应不同特性的回归数据集。为了进一步提高回归精度,基于降低回归数据复杂度的分段回归思想,在上述多尺度回归的基础上提出了双-多尺度回归,并验证了方法的有效性。最后,将提出的算法以实际的光纤陀螺温度漂移数据进行验证,结果表明,相比于传统的支持向量机和反向传播神经网络具有更好的回归精度,温度漂移模型也更加精确,以均方误差指标为例,回归精度提升了两个数量级。

Modeling and compensation method of FOG temperature drift based on multi-scale and improved support vector machine

A multi-scale modeling and regression method is proposed based on an improved support vector machine to improve the accuracy and the compensation effect of FOG (fiber optic gyroscope) temperature drift model.Firstly,the factors that cause the FOG temperature drift are analyzed,and the attribute parameters and the temperature test are given.According to the change trend of IMF (intrinsic mode function) permutation entropy obtained by empirical mode decomposition,the relationship between the regression accuracy and the entropy is given,and a multi-scale regression method based on signal decomposition is proposed.In order to improve the adaptability of multi-scale regression algorithm to different characteristics of the regression data sets,a new support vector machine (SVM) algorithm based on combined kernel function (CKF) is proposed from traditional SVM.In order to further improve the regression accuracy,a two-and multi-scale regression method is proposed by using piecewise regression to reduce the complexity of regression data,and the effectiveness of this method is verified by the test results.Finally,this two-and multi-scale CKF SVM regression algorithm is verified by simulation using actual temperature drift data of FOG,which shows that the regression accuracy is better than those of traditional SVM and BP neural network,and the temperature drift model is more accurate.Taking the MSE index as an example,the regression precision has been improved by two orders of magnitude.