Gyroscope array fusion algorithm based on deep learning and Kalman filtering

To address the challenges of noise interference and insufficient fusion accuracy in MEMS gyroscope array measurements, this study proposes a novel data fusion methodology integrating Denoising Autoencoders (DAE), Kalman Filter (KF), and BP neural networks, aiming to enhance inertial measurement precision and data stability. The framework is systematically implemented in three progressive stages: Initially, a DAE architecture is constructed to perform nonlinear denoising on raw gyroscopic signals. Subsequently, the denoised data is processed through an optimal KF algorithm to achieve preliminary fusion and generate suboptimal estimates. Ultimately, a specially designed BP neural network fusion module performs deep-level nonlinear fusion, comprehensively integrating both denoised data and KF outputs to derive global optimal estimates. Experimental validation demonstrates that the proposed algorithm achieves superior performance in both static and dynamic scenarios compared with conventional mean filtering and standalone KF approaches. Notably, the denoising capability of DAE exhibits significant enhancement in dynamic environments relative to static conditions.