基于Noise2Noise的静电悬浮惯性传感器噪声抑制方法

Noise suppression method of electrostatic suspension inertial sensor based on Noise2Noise

针对星载静电悬浮惯性传感器噪声复杂,在轨测量数据真值未知,传统方法难以有效抑制的问题,提出了基于无监督学习的Noise2Noise框架,结合集成学习方案,设计了基于Noise2Noise无监督学习的广谱随机噪声抑制方法,并基于GRACE-FO加速度计数据进行了实验验证。实验结果表明所提方法相较于传统噪声抑制方法的噪声均方误差下降8%以上,使用集成学习后,噪声水平进一步下降至12%以上。此外,所提方法在有效抑制高频噪声的同时,能够识别出高频数据中的特征性信号,可为惯性传感器载荷在轨运行状态评估提供信息保障。

In view of the noise of onboard electrostatic suspension inertial sensors is complex and difficult to be effectively suppressed by traditional methods when the true value of in-orbit measurement data is unknown,a Noise2Noise framework based on unsupervised learning is proposed.Combined with an integrated learning scheme,a broad spectrum random noise suppression method based on unsupervised learning framework of Noise2Noise is designed,which is verified by experiments based on GRACE-FO accelerometer data.The experimental results show that compared with traditional noise suppression methods,the mean square error of noise of the proposed method is reduced by more than 8%,and the noise level is further reduced by more than 12%after using integrated learning.In addition,the proposed method can effectively suppress the high-frequency noise and identify the characteristic signals in high-frequency data,which can provide information guarantee for the on-orbit operation status evaluation of inertial sensor payloads.