New modeling approach of scale-factor temperature drift based on Gaussian process regression for FOG

It is an efficient method to reduce the temperature drift error of fiber optic gyroscope (FOG) by modeling and compensation in software. Due to the combined impact of the temperature and the input angular rate, the characteristic of FOG scale factor is complicated nonlinear. To solve the problem that the traditional method can´t describe this nonlinearity and its accuracy is low, a novel approach based on Gaussian process regression (GPR) is proposed to model and compensate the scale-factor temperature drift of FOG. The approach establishes the “black-box model” which maps the temperature and the primary output of FOG to the target output of FOG, and the powerful regression ability of GPR is used to identify the nonlinear mapping relationship by learning the training data. The experiment results show that, compared with the surface regression method, the model established in the current paper can reflect the characteristic of scale factor temperature drift more accurately, and can obtain higher accuracy and better predictive ability. The training error and predicted error of the new model are less than 1 × 10^-3 o/s and their root mean square error are 6.57 × 10^-5 o/s and 1.34 × 10^-4 o/s respectively.