In order to improve the medium and long term prediction accuracy of polar motion, the LS+Informer(least-squares extrapolation and Informer, LS+Informer) prediction model is proposed based on the traditional LS+AR prediction model combined with deep learning method. The mean absolute error(MAE) was used as the evaluation index based on LS+AR forecast model and Bulletin A forecast results. After LS+Informer method is adopted, the medium and long term prediction accuracy of PM_X and PM_Y components of polar motion can be significantly improved, and the amplitude can reach 61.94% and 64.86% respectively. At the same time, the MAE value can be controlled within 10 mas in the medium and long term 365-day phase. The results show that the prediction accuracy of LS+Informer model is better than that of LS+AR model, and better than that of Bulletin A. The results demonstrate that the LS+Informer model can be effectively used in medium and long term polar motion prediction.

This paper introduces the data preprocessing and interpolation methods for universal time(UT1) and analyzes the factors affecting the interpolation accuracy of UT1, including different interpolation methods and differences in data preprocessing. The study combines international products with domestic co-observation data to analyze the interpolation accuracy of UT1 under different observation frequencies of very long baseline interferometry(VLBI). The results show that among different interpolation methods, cubic spline interpolation is superior. The interpolation process of UT1 must consider the influence of Earth's solid tides, while high-frequency corrections can be neglected. Compared with daily observations, when the observation interval is less than 2 days, the deviation of UT1 can be stabilized at around 0.05 ms. Increasing the observation interval further will lead to a more pronounced increase of UT1 deviation. Verification using domestic observational data shows that with an observation interval of 2 days, the accuracy of UT1 can reach around 0.05 ms.

To address the issues of large pseudorange measurement noise, discontinuous observation data, weak geometric strength in smartphone GNSS observations, and high noise levels in the built-in micro inertial measurement unit(MIMU) of smartphones, a tightly coupled GNSS/MIMU navigation method enhanced by low-Earth orbit(LEO) satellites is proposed. Based on the characteristics of the Walker constellation and the Keplerian orbital parameters of LEO satellites, LEO pseudorange observation data are simulated. The positioning performance of smartphone GNSS single-point positioning, GNSS/MIMU tight integration, and LEO-enhanced GNSS/MIMU tight integration are compared and analyzed. Real smartphone data are used for validation. The experimental results show that after incorporating LEO satellites, the average number of visible satellites increases by about 7. The faster movement speed of LEO satellites further optimizes the GNSS geometric strength, reducing the average PDOP value by 50%. Compared with GNSS single-point positioning, the GNSS/MIMU tight integration improves positioning accuracy in the east, north, and up directions by 18%, 23%, and 10%, respectively. The LEO-enhanced GNSS/MIMU tight integration improves positioning accuracy in the east, north, and up directions by 73%, 81%, and 60% compared with GNSS single-point positioning, and by 67%, 75%, and 56% compared with GNSS/MIMU tight integration.

The presence of differential code bias(DCB) can affect the accuracy of total electron content(TEC) calculations and pseudorange observations, thereby impacting navigation, positioning, timing, and meteorological research results.In order to accurately estimate DCB, this paper analyzes the temporal variations of GPS inter-frequency DCB data provided by the Chinese Academy of Sciences(CAS) from 2021 to 2022 and proposes an LR-KF-LSTM combined model for precise prediction and analysis of DCB. Experimental results indicate that the average absolute percentage error of this method is less than 1.9%, the average absolute error is less than 0.03 ns, and the root mean square error is less than 0.04 ns.Compared with the LSTM model, BP neural network model, and the DCB values from the CAS product, the combined model shows better prediction performance under different solar activity and geomagnetic conditions. This combined network model can effectively predict satellite DCB and also provides a reference for addressing the issue of single-day or multi-day missing data in the DCB data CAS products.

The accuracy of gravity matching navigation is influenced by the adaptability of the sailing sea area. To ensure the effectiveness of matching, it is crucial to conduct path planning for the submersible. In response to the issues of low efficiency, susceptibility to local optimal solutions, and unstable performance of existing algorithms, this paper proposes a hybrid algorithm-particle swarm optimization and grey wolf optimization(PSO-GWO) to enhance the path planning efficiency of submersibles. Building on the traditional grey wolf optimizer(GWO), the proposed algorithm incorporates the memory mechanism of the best position information of particle movement from particle swarm optimization(PSO), reconstructing the update functions for velocity and position, thereby improving the optimization capability and convergence speed of the GWO algorithm. Experiments were conducted using PSO-GWO, GWO, and PSO algorithms on a two-dimensional grid map discretized based on adaptability. The results show that compared with the GWO and PSO algorithms, the PSO-GWO algorithm reduces the planned path length by 10.99% and the number of iterations by 82.43%. This algorithm reduces redundant travel and accelerates convergence, significantly improving the underwater navigation efficiency of the submersible.

We select the hourly data of two components of extensometer at Yichang station from 2013 to 2023, calculate the M₂ wave tidal factors of the two components, perform variational mode decomposition(VMD) on them, and compute the power spectral density(PSD) for each decomposition results. Based on this, a Hilbert time-frequency analysis is conducted. The results show that NS component contains information with a frequency of 0.002 7/d, which may be related to the annual variation of water level in the Three Gorges reservoir(with a frequency of 0.002 7/d) and the current geological stress state. The EW component contains information with a frequency of 0.004-0.012/d, which may be related to the rapid changes of water level during the flood season in Three Gorges reservoir(with frequencies of 0.004-0.007/d and 0.010-0.013/d). Combining the VMD decomposition results of tidal factors and earthquake events with magnitude above 4.0 within 200 km of Yichang station, abnormal high value is observed in the detailed information of EW component tidal factor before the Badong M5.1 earthquake in 2013. Additionally, in terms of meteorological factors, the tidal factors do not exhibit a regular annual variation similar to temperature. However, trend changes in air temperature lead to cave temperature changes, which has a certain impact on the variation trend of tidal factors.

The linear or nonlin drift of the spring gravimeter must be addressed for the existing observation system based on spring gravimeters to monitor the long-term change signal of gravity. Simple linear fitting is usually difficult to simulate the real drift of the instrument over a long period of time, and too high-order fitting will also remove the trend change signal of the signal. This study introduces linear and nonlinear drift correction models to more reasonably remove the drift. This study uses the data of the gPhone gravimeter at Zigui station in 2017 and the gPhone gravimeter at Wushi station in 2023 for analysis and verification. First, the correlation between each channel of Zigui station and the original gravity channel is analyzed, and the theoretical gravity effect of the Three Gorges water storage on the Zigui station is calculated with the results that reach more than 100 μGal. Then, the model coefficients are obtained using a variety of regression analysis methods, and the residual results are studied after deducting various correction values and drift terms. The research results show that the changes in the internal pressure and temperature of the instrument are the main causes of gravity data drift. In the regression model used in this paper, the nonlinear regression model(pure quadratic regression) drift removal results can better reflect the gravity changes caused by water storage, and the residual effect obtained after smoothing is closer to theory than the conventional polynomial fitting method for drift removal. Similar conclusions were obtained by verifying the data from Wushi station.

We conduct experiments on the parallel operation of DC and AC geoelectrical resistivity observation systems at Chengdu earthquake monitoring center station by setting different power supply parameters, and summarize the interference characteristics of different geoelectrical resistivity observation systems on geomagnetic records at the same station. The results show that the interference caused by DC geoelectrical resistivity observation on geomagnetic records is in the form of positive and negative large amplitude square waves, and the output current amplitude is positively correlated with the interference amplitude. The interference caused by AC geoelectrical resistivity observation on geomagnetic records is in the form of large amplitude sine waves, and will not interfere geomagnetic records when the output current frequency is 1-10 Hz and the output current amplitude is 0.1-0.5 A. In addition, based on the experimental results of AC geoelectrical resistivity observation, we simulate the amplitude frequency characteristic curve of the FGM-01 fluxgate magnetometer, which conforms to the frequency response characteristics of a low-pass filter.

Based on the earthquakes recorded by Gansu early warning system since its operation and the output results of early warning system, a comparative study is conducted on the output magnitude of earthquake early warning system and the officially cataloged magnitude, and the distribution of magnitude errors is analyzed. We use the back-propagation neural network to learn and predict the warning magnitude parameters, and obtain the warning magnitude. The results indicate that for earthquakes with smaller actual magnitudes, the estimated magnitude is larger, while for earthquakes with larger magnitudes, the estimated magnitude is smaller. The neural network propagation algorithms can simulate and predict warning magnitude, and reduce the deviation between warning magnitude and cataloged magnitude.

An adaptive method for removing power frequency interference is proposed by combining seasonal trend dispersion(STD) decomposition with recursive least squares(RLS). This method utilizes STD decomposition to extract the seasonal component containing power frequency interference from microseismic waves, which serves as the reference signal for RLS, and dynamically updates the algorithm coefficients to make the calculated signal approach the power frequency interference signal. Simulation experiments were designed by superimposing 1 set of microseismic waves without power frequency interference with 3 different types of power frequency interference. The finite impulse response(FIR) filtering, wavelet threshold(WT) filtering, and the method presented in this paper were used for processing. The results indicate that the method in this paper can effectively remove power frequency interference while fully preserving the key time-frequency characteristics of microseismic waves. Furthermore, the application of this method to the microseismic monitoring data processing of Taoyuan coal mine has verified its feasibility and reliability in engineering applications.