Modeling Temperature Data of RLG's Scale Factor Using LS-SVM

In this paper, a LS-SVM model based RLG's scale factor temperature data modeling method is studied. Using traditional least square linear model to modeling nonlinear ring laser gyro scale factor test data has its intrinsic shortcoming, and sometimes it is difficult to meet the application requirements. Recently, nonlinear function approximation based modeling methods such as the BP networks are introduced to modeling the temperature data. But in the BP networks will suffer the problem of overfitting and the existence of many local minima. To avoid these shortcomings the LS-SVM is used to modeling the scale factor temperature data. Base on the analysis of the test scale factor data, the scale factor test data is modeled as the function of the temperature and its increment, and LS-SVM model is employed to estimate the nonlinear function. The simulation results show that the LS-SVM model can approach scale factor data accurately, and its precision is much higher than the least square model. The mean squared deviation of LS-SVM model is smaller than 0.51times10
^-6
("/pulse). Base on considerately designed test procedure and large numbers of experimental data, a practical temperature model can be established.