Temperature Drift Compensation Method for Fiber-Optic Gyroscope Based on LSTM-PINN

Addressing the core issue of fiber optic gyroscope accuracy and stability being constrained by temperature drift, this paper proposes a compensation method based on a hybrid model combining Long Short-Term Memory (LSTM) and Physical Information Neural Network (PINN). This method employs LSTM to perform high-precision modeling of the time-series dynamic characteristics of the gyroscope's output during temperature changes. Additionally, physical constraints are introduced within the neural network framework to enhance the model's interpretability and physical consistency, effectively suppressing output deviations caused by environmental temperature disturbances. To demonstrate the generalizability of the new method, training data was selected from a temperature range of -10°C to 30°C, while test data was expanded to a more stringent range of -20°C to 50°C. The compensation results indicate: Compared to other methods, the proposed LSTM-PINN model demonstrates significant performance improvements, with the standard deviation (STD), mean absolute error (MAE), and root mean square error (RMSE) of the output signals reduced by up to 35.7%, 41.1%, and 34.5%, respectively, providing an effective solution for enhancing the environmental adaptability of fiber optic gyroscopes.