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

Area of Science:

• Instrumentation
• Signal Processing
• Machine Learning

Background:

• Temperature-induced bias drift is a critical issue affecting fiber optic gyroscope (FOG) accuracy.
• Current compensation techniques struggle to balance fitting accuracy with generalization capabilities.

Purpose of the Study:

• To develop an advanced algorithm for effective temperature compensation in FOGs.
• To overcome the limitations of existing polynomial fitting and deep learning methods.

Main Methods:

• A novel physics-informed random forest (PI-RF) algorithm was developed.
• Integration of a temporal memory vector (multi-scale sliding window) and a temperature-material dynamic modulation vector.
• Adaptive hyperparameter optimization was employed to enhance algorithm performance.

Main Results:

• The PI-RF algorithm significantly improved bias stability from 0.282 to 0.006°/h.
• Achieved an 89.15% optimization in bias instability.
• Demonstrated enhanced physical interpretability, generalization, and extrapolation capabilities.

Conclusions:

• The proposed PI-RF algorithm offers a feasible and effective solution for FOG temperature compensation.
• This approach enhances gyroscope accuracy and reliability in varying temperature conditions.
• The method provides a promising direction for future research in inertial sensor compensation.