The stability of absolute gravimeters during carriage-falling body separation is crucial for improving gravitational acceleration measurement accuracy. Transmission speed accuracy of the transmission system and system vibration are core factors determining this stability, while steel belt pre-tightening force, free-fall segment acceleration, and start-up segment displacement are key parameters influencing both. In-depth analysis of their coupling clarified their roles, and two objective function models (for speed accuracy and vibration) were established, with fitting accuracies R² = 0.8976 and R² = 0.8395, respectively. Since traditional single-objective optimization fails to balance “improving speed accuracy” and “suppressing vibration”, this study proposes a multi-objective optimization method: two Nondominated Sorting Genetic Algorithm II (NSGA-II) parameter sets were designed, Hypervolume (HV) index quantified solution set quality, and Wilcoxon signed-rank test was combined to determine the optimal parameter set; comparing the Global Criterion Method and Weighted Sum Method, the former was superior (no dimensional bias) and more suitable for this study, finally screening out the optimal parameter combination. Experimental results showed that the measured transmission speed accuracy was 0.09132 m/s (16.94% lower than the orthogonal experiment’s optimal level); the measured system vibration was 0.022 m/s², falling within the orthogonal experiment’s optimal range. Consequently, separation moment stability was significantly enhanced, with its standard deviation reduced by 45% pre-optimization. This method achieves global balance in transmission system dynamic performance, providing an effective parameter optimization strategy for improving absolute gravimeter measurement accuracy.

To ensure the sustained stability of absolute gravity benchmark points from 2017 to 2024, observational records from superconducting gravimeters (SGs) and absolute gravimeters were comprehensively examined in this work, and the environmental effects on gravitational acceleration were quantitatively assessed. The annual fluctuation of the SG (iGrav-012k) scale factor reached 0.268 μGal/V, with a weighted average of (–92.8702 ± 0.0265) μGal/V (relative precision of 0.3‰), providing a precise scale factor for long-term SG monitoring. By removing step discontinuities in the SG data using FG5-X249 absolute gravimeter measurements, the residual fitting error decreased to 6.3 μGal. In addition, the SG drift was estimated as 1.0 μGal/year through international comparison datasets and FG5 measurements, substantially improving the consistency of the time series. Further investigation showed that the SG residuals exhibited clear seasonal oscillations, which were mainly attributed to local hydrological processes and ground deformation near the benchmark sites. By integrating groundwater level and deformation monitoring data and applying a neural network model to separate hydrological load components, the peak-to-peak residual amplitude was reduced from 13 μGal to 3.5 μGal. Quantitative analysis indicated that hydrological effects contributed about 9.5 μGal to the seasonal variation, whereas surface deformation had only a minor impact (<2 μGal). The findings confirm that careful data correction and isolation of environmental effects are effective for maintaining the long-term stability of gravity benchmarks. The developed workflow provides a reproducible framework for high-precision gravity site maintenance and supports future dynamic monitoring of regional environmental load responses.