The Study Of Nonlinear Dynamic Modeling Method In Accelerometer

The dynamic model of accelerometer is one of the main methods to describe the dynamic characteristics of accelerometer. Researching on dynamic characteristics of the accelerometer, it is that often ignore the nonlinear characteristics of accelerometer and the accelerometer is equivalent to a linear second-order system of spring-damper-quality. However in the actual test environment, due to the changes of temperature and humidity as well as other factors, there is always some nonlinear characteristics. In this time, the accelerometers for linear will not accurately describe the dynamic characteristics, the second-order linear model is established according to the physical structure of the accelerometer has been unable to meet the measurement requirements in complex environment. Therefore, using the input and output data of accelerometer calibration experiment to establish a nonlinear dynamic model of accelerometer has become an important content of current research on accelerometer.This paper studies the advantages and disadvantages of the local kernel and global kernel in SVM (Support Vector Machine, SVM) learning algorithm. The weak learning defects and poor generalization ability of using the single kernel is effectively overcame in dealing with nonlinear problems by constructing a mixed kernel. The second-order linear spring-quality-damper model of accelerometer is introduced. The nonlinear dynamic modeling method based on mixed kernel SVM is proposed. By building mixed kernel from the linear kernel and RBF kernel function composition, using the dynamic calibration test data of accelerometer, using Akaike information criterion to determine the model order, by adjusting the weighting factor of mixed kernel, the nonlinear dynamic model of accelerometer is modeling.Experiments show that the mixed kernel function is combined by linear kernel and RBF kernel function. The accelerometer model is divided into linear and nonlinear parts connected in series. The nonlinear dynamic model of accelerometer is modeling. Comparing with the second-order model and single kernel function SVM model, the results show that the nonlinear dynamic modeling method of accelerometer based on mixed kernel SVM is effective and the established model of accelerometer than the traditional second-order linear model and single kernel function model has higher simulation precision and better generalization performance.