Physics-informed random forest algorithm for temperature compensation in fiber optic gyroscopes

Temperature-induced bias drift severely degrades the accuracy of fiber optic gyroscopes. Existing compensation methods face a dilemma between polynomial fitting accuracy and deep learning generalization. To solve this, a physics-informed random forest (PI-RF) algorithm is proposed. By integrating a temporal memory vector based on a multi-scale sliding window and a temperature-material dynamic modulation vector, together with adaptive hyperparameter optimization, the PI-RF algorithm effectively improves physical interpretability, generalization, and extrapolation capabilities. Experiments demonstrate that the PI-RF algorithm improves bias stability from 0.282 to 0.006°/h, optimizing bias instability by 89.15%. This work provides a feasible solution for temperature compensation of gyroscopes.