Efficient modeling of fiber optic gyroscope drift using improved EEMD and extreme learning machine

• We propose method to model FOG drift under dramatic temperature variation.
• We propose a novel criterion for selecting relevant modes of EMD.
• We verify that extreme learning machine can model FOG short term drift.
• We establish new prediction structure for complicate FOG signal.

In order to model the drift of fiber optic gyroscope (FOG) efficiently, a novel multi-scale prediction method is proposed by utilizing signal decomposition. Analytical expression of thermally induced drift of FOG is given first, which forms our theoretical basis of multi-scale prediction. Newly proposed bounded EEMD is used to decompose drift signal into a series of stationary modes, and then an adaptive feature selection criterion is proposed to construct distinct sub-series. Extreme learning machine is used to train these sub-series respectively, and a hybrid model is then obtained by adding up all the sub-models. Experiments have shown that, compared with the state-of-the-art methods, the proposed method improves prediction accuracy by two orders and achieves much faster speed in training process.