Temperature Compensation for Fiber Optic Gyroscope Based on Bayesian Optimized Relevance Vector Regression

Fiber optic gyroscope (FOG) has been widely utilized in navigation, guidance, orientation, stabilization and related applications due to its potential advantages. However, thermal bias drift induced by temperature variation often seriously deteriorates its operational accuracy. A data driven temperature compensation method based on Bayesian optimized relevance vector regression is proposed to reduce thermal bias drift. Datasets with temperature and temperature derivative as input samples and thermal bias drift as output samples were constructed from full temperature experiments, and a data driven sparse probabilistic relevance vector regression model with Bayesian optimized hyper-parameters were trained, and used to reduce thermal bias drift of the FOG. Temperature compensation experiments demonstrate that using the proposed method, thermal bias drift can be reduced approximately to comparable accuracy level as at constant room temperature, thus greatly promote adaptability of the FOG towards different temperature environments.