Drift compensation is crucial for micro-thrust measurement equipment to detect gravitational waves as ultra-low frequency drift poses a significant challenge. This study proposes a drift compensation method taking advantages of predictions using pre- and post-instrument heating signals. We de-composite the signals into baseline drifts and alternating noise. It is demonstrated that the baseline drift could be fitted through improved weighted least squares fitting. Besides, the alternating noise can be decomposed using wavelet method and predicted using a novel lightweight Seg-Mamba model. The results illustrated that the proposed method can reduce power spectral density (PSD) of background signal drift from 49930 μN/Hz^1/2 to 0.08 μN/Hz^1/2 at 0.1 mHz for micro thrust measurement. The proposed method is capable of effective drift compensation for thrust measurement in complex and dynamic environments.

Micro-thrust measurement, drift compensation, ultra-low frequency, mamba, time series prediction