Temperature drift modeling and compensation of accelerometer applied in atom gravimeter

Vibration correction by using an accelerometer to measure vibration is one of the critical technologies for field measurement of atom gravimeters. However, the accelerometer is often affected by temperature and other environmental factors, resulting in a drift of vibration signal. Temperature drift of accelerometer modeling and compensation is one of the main methods to restrain the drift. Considering the different effects on the drift of various temperatures, temperature change rates, and temperature rise and drop processes, taking temperature and temperature change rate as inputs, we establish the temperature drift model of the accelerometer and use the positive and negative sign of temperature change rate to indicate the rise and drop of temperature. The model is described by support vector regression (SVR), and the particle swarm optimization (PSO) algorithm is utilized for tuning the model parameters. We carried out the heating and cooling experiments with temperature change rates of ±2 °C/min, ±1 °C/min, and ±0.5 °C/min, respectively. Instead of modeling separately for various temperature change rates, we trained the same PSO-SVR model with the accelerometer drift data obtained under various temperature conditions to improve accuracy and adaptability. By comparing its accuracy with the PSO tuned back-propagation neural network and polynomial least squares models and their performance on non-training data, we verified that the PSO-SVR model has a good compensation effect and excellent generalization performance, and the atom gravimeter interferometric phase noise induced by vibration is significantly reduced after compensation. The method proposed in this paper is of reference significance for improving the field measurement accuracy of the atom gravimeter.