Geological Disaster Monitoring Method and Device Based on BeiDou GNSS Reference Station for Short-Term Rainfall Early Warning

By integrating meteorological sensors and anti-interference quality control into the BeiDou GNSS reference station, and utilizing existing resources for atmospheric precipitable water conversion and early warning, the problems of unused water vapor information and false alarms due to signal interference in geological disaster monitoring by the BeiDou GNSS reference station have been solved, achieving efficient and reliable short-term rainfall early warning and geological disaster monitoring.

CN123066892APending Publication Date: 2026-11-10WUHAN MEASURING FUTURE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611225304.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing BeiDou GNSS reference stations have failed to effectively utilize water vapor information in geological disaster monitoring. New facilities are costly and prone to false alarms due to signal interference, making large-scale application difficult.

Method used

By reusing the radio frequency links and data interfaces of existing BeiDou GNSS reference stations and combining meteorological sensor data to perform precise single-point positioning calculations, the zenith wet delay is obtained and converted into atmospheric precipitable water. After anti-interference quality control, the data is pushed to the cloud AI model for short-term rainfall warnings and to generate geological disaster monitoring and warnings.

Benefits of technology

It has achieved low-cost and high-reliability geological disaster monitoring and early warning, effectively utilized existing resources, reduced false alarms caused by signal interference, and improved the accuracy and efficiency of monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123066892A_ABST
    Figure CN123066892A_ABST
Patent Text Reader

Abstract

This invention provides a geological disaster monitoring method and device based on short-term rainfall early warning using a BeiDou GNSS reference station. The method includes the following steps: acquiring raw observation data streams and collecting meteorological sensor data from an added integrated extension module, uniformly registering them to the same time reference; calculating a weighted average temperature based on temperature and relative humidity, converting the zenith wet delay into atmospheric precipitable water content, and generating a time-series data package of atmospheric precipitable water content; performing anti-interference quality control on the atmospheric precipitable water content time-series data package, identifying anomalous epochs based on signal quality indicators, and performing time-series repair on the atmospheric precipitable water content of anomalous epochs; outputting short-term rainfall early warning results for a preset future period through a cloud-based AI rainfall prediction model and monitoring geological disasters. By reusing existing reference stations and employing edge-cloud collaborative anti-interference, the economic efficiency and reliability of geological disaster monitoring and early warning deployment are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of meteorological monitoring and disaster early warning technology, and in particular to a geological disaster monitoring method and device based on short-term rainfall early warning using BeiDou GNSS reference stations. Background Technology

[0002] A large number of BeiDou GNSS reference stations have been built across the country for deformation monitoring, displacement monitoring, and CORS positioning. Taking geological disaster monitoring as an example, tens of thousands of BeiDou GNSS monitoring stations have been deployed in areas prone to landslides, collapses, and ground subsidence. These stations acquire millimeter-level position change information through carrier phase differential or precise point positioning technology to determine surface deformation trends and disaster risks. At the same time, a large number of CORS reference stations for mapping and positioning have also been built in various regions, forming a high-density BeiDou GNSS continuous observation network covering most parts of the country.

[0003] Currently, the original purpose of the aforementioned BeiDou GNSS reference stations only utilizes the geometric calculation results of the carrier phase. Specifically, deformation monitoring and CORS positioning only focus on the time-series changes of the station coordinates, extracting displacement or position information by performing differential or precise single-point positioning calculations on the carrier phase observations. However, when BeiDou GNSS signals traverse the atmosphere, their propagation path is affected by tropospheric delay—this delay carries rich atmospheric water vapor information. Specifically, BeiDou GNSS signals undergo refraction and delay when propagating in the troposphere. The delay caused by the dry air components in the atmosphere is called zenith dry delay, and the delay caused by water vapor is called zenith wet delay. Zenith wet delay is closely related to the water vapor content in the atmosphere, and can be converted to obtain the atmospheric precipitable water content. In existing deformation monitoring and CORS positioning applications, tropospheric delay is usually only treated as an estimated parameter or error source in positioning calculations, and the water vapor information it contains is not extracted and utilized.

[0004] To add areal rainfall monitoring capabilities to the existing BeiDou GNSS monitoring network coverage area, one direct approach is to build a new, independent BeiDou GNSS areal rainfall meter—relocating the site, erecting poles, installing receivers and meteorological sensors, and deploying power and communication facilities. This method is costly, time-consuming, and faces challenges such as site selection difficulties and poor foundation construction conditions in some geologically hazardous areas, making large-scale deployment difficult in the short term.

[0005] Furthermore, installing water vapor inversion equipment near existing BeiDou GNSS reference stations presents another technical challenge: geologically hazardous areas are often located in mountainous regions or areas with complex electromagnetic environments. Interference sources such as lightning, high-voltage power lines, and communication base stations can degrade the quality of BeiDou GNSS signals, causing drift in the total zenith delay (PWV) calculation. If the interfered PWV data is directly input into the AI ​​rainfall prediction model, the model may misjudge signal anomalies as water vapor anomalies, resulting in false alarms for heavy rain. Existing maintenance methods lack effective interference identification and location mechanisms. Once a PWV anomaly occurs, maintenance personnel find it difficult to quickly distinguish whether it is caused by BeiDou GNSS signal interference, equipment failure, or changes in meteorological conditions, leading to low troubleshooting efficiency and high maintenance costs.

[0006] Therefore, the key issue restricting the large-scale application of BeiDou GNSS areal rainfall prediction technology is how to reuse the hardware resources of existing BeiDou GNSS reference stations at the lowest cost without building new independent facilities, so that they can simultaneously have the ability to monitor geological disaster displacement and meteorological cause water vapor, and ensure the reliability of PWV output when BeiDou GNSS signals are interfered with. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a geological disaster monitoring method based on short-term rainfall early warning using BeiDou GNSS reference stations, to solve the technical problems of high construction costs, unutilized existing resources, false alarms due to interference, and difficulty in tracing faults in the prior art. The method includes:

[0008] The BeiDou GNSS raw observation data stream is obtained from the radio frequency link or data interface of the existing BeiDou GNSS reference station, and meteorological sensor data is collected from the added integrated extension module. The BeiDou GNSS raw observation data stream and the meteorological sensor data are uniformly registered to the same time reference. The BeiDou GNSS raw observation data stream includes carrier phase, pseudorange, Doppler, signal-to-noise ratio and observation timestamp, and the meteorological sensor data includes air pressure, temperature and relative humidity. At the edge end, precise single-point positioning calculation is performed on the raw BeiDou GNSS observation data stream to obtain the total zenith delay. The zenith dry delay is estimated based on the air pressure and the station coordinates of the existing BeiDou GNSS reference stations. The zenith wet delay is obtained by subtracting the zenith dry delay from the total zenith delay. The weighted average temperature is calculated based on the temperature and the relative humidity. The zenith wet delay is converted into atmospheric precipitable water content, and atmospheric precipitable water content time-series data packets are generated. Anti-interference quality control is performed on the atmospheric precipitable water time series data packet. The signal quality index of each epoch is calculated in real time. Abnormal epochs are identified based on the signal quality index, and the atmospheric precipitable water in the abnormal epochs is time-series repaired to generate atmospheric precipitable water time series data packets with quality tags. The time-series data package of atmospheric precipitable water, which has been marked with quality, is pushed to the cloud-based AI rainfall prediction model. The cloud-based AI rainfall prediction model outputs short-term rainfall warning results for a preset period in the future, and determines the trigger risk level of geological disasters based on the short-term rainfall warning results, so as to monitor geological disasters.

[0009] This invention also provides a geological disaster monitoring device based on short-term rainfall early warning using a BeiDou GNSS reference station, to solve the technical problems of high construction costs, unutilized existing resources, false alarms due to interference, and difficulty in tracing faults in the prior art. The device includes: The original BeiDou GNSS antenna and the original BeiDou GNSS receiver of the existing BeiDou GNSS reference station, wherein the original BeiDou GNSS antenna is used to receive BeiDou GNSS radio frequency signals, and the original BeiDou GNSS receiver is used to perform the original displacement monitoring service; A broadband radio frequency power divider, whose input is connected to the original BeiDou GNSS antenna, and whose output is divided into a main path and a branch path. The main path is connected to the original BeiDou GNSS receiver, and the branch path is connected to a low-noise amplifier. Low-noise amplifiers are used to compensate for the gain of branch radio frequency signals; The multi-frequency BeiDou receiver module is connected to the output of the low-noise amplifier and is used to capture, track and demodulate the branch radio frequency signal after gain compensation to generate a BeiDou GNSS raw observation data stream. The integrated expansion module integrates the multi-frequency Beidou receiver module, temperature, humidity and pressure sensor, rain gauge interface, edge computing MCU and 4G / Beidou short message dual communication module. It is used to collect meteorological sensor data and perform precise single-point positioning calculation, atmospheric precipitable water inversion and anti-interference quality control at the edge, and generate atmospheric precipitable water time series data packets with quality tags. A power management module, including a solar panel and a battery expansion kit, is used to power the integrated expansion module; The rainfall prediction module is used to push the quality-tagged atmospheric precipitable water time series data package to the cloud AI rainfall prediction model, and output the short-term rainfall warning results for the future preset period through the cloud AI rainfall prediction model. The short-term rainfall warning results are used for geological disaster monitoring and early warning. The comprehensive geological disaster assessment module is used to generate comprehensive geological disaster early warning information when the risk level of the rainfall trigger indicated by the short-term rainfall warning result exceeds a preset risk threshold and the geological disaster deformation trend determined according to the original displacement monitoring data intensifies.

[0010] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least: By reusing the radio frequency links and data interfaces of existing BeiDou GNSS reference stations to obtain BeiDou GNSS observations, and combining precise single-point positioning calculations with anti-interference quality control, reliable short-term rainfall warning results are obtained, thus achieving low-cost and high-reliability geological disaster monitoring and early warning. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a geological disaster monitoring method based on short-term rainfall early warning using a BeiDou GNSS reference station, provided by an embodiment of the present invention. Figure 2 This is a structural diagram of the existing BeiDou GNSS reference station retrofit device according to an embodiment of the present invention; Figure 3 This is a flowchart of the anti-interference PWV inversion method according to an embodiment of the present invention; Figure 4 This is a flowchart of the PWV dual-channel BeiDou GNSS data acquisition and selection process according to an embodiment of the present invention; Figure 5 This is a flowchart of the dual-channel selection process according to an embodiment of the present invention; Figure 6 This is a flowchart of the end-to-cloud consistency verification process according to an embodiment of the present invention; Figure 7 This is a flowchart illustrating the overall modification process of an embodiment of the present invention. Detailed Implementation

[0013] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0014] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] In this embodiment of the invention, a geological disaster monitoring method based on short-term rainfall early warning using BeiDou GNSS reference stations is provided, such as... Figure 1 As shown, the method includes: Step S101: Obtain the raw BeiDou GNSS observation data stream from the radio frequency link or data interface of the existing BeiDou GNSS reference station, and collect meteorological sensor data from the added integrated extension module. Register the raw BeiDou GNSS observation data stream and the meteorological sensor data to the same time reference. The raw BeiDou GNSS observation data stream includes carrier phase, pseudorange, Doppler, signal-to-noise ratio and observation timestamp. The meteorological sensor data includes air pressure, temperature and relative humidity. Step S102: Perform precise single-point positioning calculation on the BeiDou GNSS raw observation data stream at the edge end to obtain the total zenith delay, and estimate the zenith dry delay based on the air pressure and the station coordinates of the existing BeiDou GNSS reference stations. Subtract the zenith dry delay from the total zenith delay to obtain the zenith wet delay. Calculate the weighted average temperature based on the temperature and the relative humidity, convert the zenith wet delay into atmospheric precipitable water, and generate atmospheric precipitable water time series data packets. Step S103: Perform anti-interference quality control on the atmospheric precipitable water time series data packet, calculate the signal quality index of each epoch in real time, identify abnormal epochs based on the signal quality index, and perform time series repair on the atmospheric precipitable water in the abnormal epochs to generate quality-tagged atmospheric precipitable water time series data packets. Step S104: Push the quality-tagged atmospheric precipitable water time series data package to the cloud AI rainfall prediction model. The cloud AI rainfall prediction model outputs short-term rainfall warning results for a preset period in the future, and determines the trigger risk level of rainfall-induced geological disasters based on the short-term rainfall warning results, so as to monitor geological disasters.

[0016] In practice, the following steps are used to obtain the raw BeiDou GNSS observation data stream from the radio frequency link or data interface of existing BeiDou GNSS reference stations: The system acquires the availability status of the original protocol and the impact level of the original displacement monitoring service for the existing BeiDou GNSS reference stations. It then determines the availability status of the original protocol. If the protocol is not open or cannot stably output raw observations, a radio frequency power division multiplexing (RF Power Division Multiplexing) data channel is selected as the target data channel, and the BeiDou GNSS raw observation data stream is acquired via the RF Power Division Multiplexing data channel. If the protocol is open and can stably output raw observations, the system determines the impact level of the original displacement monitoring service. If the determination result indicates that accessing the data forwarding channel will affect the original displacement monitoring service, the RF Power Division Multiplexing data channel is selected as the target data channel, and the BeiDou GNSS raw observation data stream is acquired via the RF Power Division Multiplexing data channel. If the determination result indicates that accessing the data forwarding channel does not affect the original displacement monitoring service, a serial port / network data forwarding channel is selected as the target data channel, and the BeiDou GNSS raw observation data stream is acquired via the serial port / network data forwarding channel.

[0017] In specific implementation, the acquisition of the BeiDou GNSS raw observation data stream via the radio frequency power division multiplexing data channel is achieved through the following steps: The BeiDou GNSS radio frequency signal output from the original BeiDou GNSS antenna feeder of the existing BeiDou GNSS reference station is input into a broadband radio frequency power divider. The broadband radio frequency power divider splits the signal into a main radio frequency signal and a branch radio frequency signal. The original BeiDou GNSS antenna feeder is the signal transmission feeder between the original BeiDou GNSS antenna and the original BeiDou GNSS receiver of the existing BeiDou GNSS reference station. The main radio frequency signal is then fed into the original BeiDou GNSS receiver. The branch radio frequency signal is then fed into a newly added multi-frequency BeiDou receiving module after gain compensation by a low-noise amplifier. The newly added multi-frequency BeiDou receiving module acquires, tracks, and demodulates the gain-compensated branch radio frequency signal to generate the original BeiDou GNSS observation data stream.

[0018] In specific implementation, the branch radio frequency signal is subjected to gain compensation by a low-noise amplifier through the following steps: The system samples the actual signal-to-noise ratio (SNR) of each BeiDou GNSS frequency point in real time and compares the actual SNR with a preset target SNR range. When the actual SNR is lower than the lower limit of the target SNR range, the gain step of the low-noise amplifier is gradually increased until the actual SNR recovers to the target SNR range. During the gradual increase of the gain step of the low-noise amplifier, the edge computing MCU synchronously monitors the noise figure and signal distortion index of each frequency point. When the noise figure exceeds a preset noise tolerance or the signal distortion index exceeds a preset distortion threshold, the gain of the low-noise amplifier is adjusted back to the step value before the current gain increase. When the actual SNR remains lower than the lower limit of the target SNR range and the gain of the low-noise amplifier has been adjusted to the maximum allowable gain but still cannot recover, an RF link anomaly flag is generated and a channel switching suggestion is reported to the cloud.

[0019] In specific implementation, the following steps are used to obtain the raw BeiDou GNSS observation data stream via the serial port / network data forwarding channel: The original BeiDou GNSS receiver connected to the existing BeiDou GNSS reference station receives RINEX observation files, RTCM streams, or manufacturer binary observation streams output by the original BeiDou GNSS receiver through the data output interface and performs protocol parsing. For each epoch after protocol parsing, a field integrity check is performed to confirm whether each epoch contains carrier phase, pseudorange, Doppler, signal-to-noise ratio, cycle slip identifier, and observation timestamp. If the field integrity check passes, the observation time of each epoch is unified to the same time base, and observation data from different manufacturers are converted into a unified internal format to generate the original BeiDou GNSS observation data stream.

[0020] In specific implementation, the following steps are used to perform anti-interference quality control on the atmospheric precipitable water time-series data packet: real-time calculation of the signal quality index for each epoch; identification of abnormal epochs based on the signal quality index; and time-series repair of the atmospheric precipitable water in the abnormal epochs to generate a quality-tagged atmospheric precipitable water time-series data packet: The system calculates the effective satellite count, position accuracy factor, signal-to-noise ratio, cycle slip rate, and precise single-point positioning solution residual for each epoch in real time, and uses these as the signal quality indicators. When the effective satellite count is less than a preset threshold, the signal-to-noise ratio drops by more than three times the mean of the regional baseline, or the cycle slip rate exceeds a preset cycle slip threshold, the current epoch is marked as an anomalous epoch. Empirical mode decomposition or Kalman filtering is used, with surface saturated vapor pressure from the meteorological sensor data as an auxiliary constraint, to perform time-series repair on the atmospheric precipitable water of the anomalous epoch, generating a repaired epoch. When the duration of continuous anomalous events exceeds a preset time threshold, the push of atmospheric precipitable water data to the cloud-based AI rainfall prediction model is stopped, and a data missing marker is generated and uploaded to the cloud. The repaired epoch, the epoch with the data missing marker, and the normal epochs not marked as anomalous are merged to generate the quality-marked atmospheric precipitable water time-series data package.

[0021] In practice, the following steps are used to format the data and initialize the model after generating the atmospheric precipitable water time series data package: The integrated extension module is registered to a cloud AI computing platform. A pre-trained Transformer rainfall prediction model matching the region of the existing BeiDou GNSS reference station is downloaded from the cloud AI computing platform. Based on the historical atmospheric precipitable water time-series data and historical rainfall data of the existing BeiDou GNSS reference station, the Transformer rainfall prediction model is adjusted to generate an adjusted Transformer rainfall prediction model. The atmospheric precipitable water time-series data package is aggregated into multiple time resolution levels at preset intervals, and data channel type tags and quality control tags are added to each epoch after aggregation to generate formatted and quality-tagged atmospheric precipitable water time-series data packages. The quality control tags include the number of effective satellites, position accuracy factor, signal-to-noise ratio, cycle slip rate, precise single-point positioning solution residual, and abnormal epoch tags. The data channel type tags are used to identify whether the current atmospheric precipitable water data comes from a radio frequency power division multiplexing data channel or a serial port / network data forwarding channel. The formatted and quality-tagged atmospheric precipitable water time-series data package is input into the adjusted Transformer rainfall prediction model, and the reliability of the input data is dynamically weighted.

[0022] In practice, the following steps are used to achieve end-to-cloud consistency verification: The edge device uploads the raw BeiDou GNSS observation data stream, the intermediate results of the precise point positioning calculation, the calculation results of atmospheric precipitable water content, and the signal quality indicators to the cloud. The cloud, based on the raw BeiDou GNSS observation data stream, uses the same precise point positioning calculation method and the same atmospheric precipitable water content retrieval method as the edge device to verify the data, obtaining the cloud-verified atmospheric precipitable water content. The cloud compares the atmospheric precipitable water content calculation results of the edge device with the cloud-verified atmospheric precipitable water content within a sliding window, calculating the difference in atmospheric precipitable water content. If the difference in atmospheric precipitable water content exceeds a preset threshold within K consecutive epochs, a consistency anomaly alarm is triggered. After triggering the consistency anomaly alarm, if both the atmospheric precipitable water content of the edge device and the cloud-verified atmospheric precipitable water content are abnormal, and the signal-to-noise ratio decreases, the cycle slip rate increases, and the position accuracy factor deteriorates, then it is determined to be BeiDou GNSS signal interference. If the atmospheric precipitable water content verified by the cloud is normal, but the atmospheric precipitable water content deviation at the edge continuously exceeds the preset threshold, it is determined to be an edge module failure; if timestamp misalignment, field missing, satellite number abnormality, or observation frequency mapping error occurs when acquiring data through the serial port / network data forwarding channel of the data interface, it is determined to be a protocol parsing error; if the zenith wet delay calculated by the edge is consistent with the zenith wet delay obtained by the cloud verification, but the atmospheric precipitable water content deviation is abnormal and the air pressure, temperature, or relative humidity changes abruptly, it is determined to be a meteorological sensor abnormality; the BeiDou GNSS signal interference, the edge module failure, the protocol parsing error, and the meteorological sensor abnormality are taken as abnormality cause types, and corresponding operation and maintenance suggestions are output according to the abnormality cause type. The operation and maintenance suggestions include channel switching suggestions, module restart suggestions, feeder or low noise amplifier inspection suggestions, meteorological sensor calibration suggestions, or original manufacturer protocol adaptation suggestions.

[0023] In practice, the monitoring of geological disasters based on the short-term rainfall warning results is achieved through the following steps: The geological hazards include at least one of landslides, collapses, and ground subsidence. After obtaining the short-term rainfall warning results, rainfall amount and intensity information for a future preset time period are extracted from the short-term rainfall warning results. The trigger risk level of the rainfall-induced geological hazards is determined based on the rainfall amount and intensity information. Original displacement monitoring data is obtained from the existing BeiDou GNSS reference stations. The deformation trend of the current geological hazard is judged based on the original displacement monitoring data to determine whether it is intensifying, and a deformation trend judgment result is generated. When the trigger risk level exceeds a preset risk threshold and the deformation trend judgment result is intensifying, the short-term rainfall warning results and the original displacement monitoring data are input into the comprehensive geological hazard assessment model to generate comprehensive geological hazard warning information.

[0024] Example 1: A Geological Disaster Monitoring Method Based on Short-Term Rainfall Early Warning Using BeiDou GNSS Reference Stations This embodiment provides a geological disaster monitoring method based on short-term rainfall early warning using BeiDou GNSS reference stations. The existing BeiDou GNSS reference stations are those already built for deformation monitoring, displacement monitoring, or CORS positioning.

[0025] (I) Data Acquisition and Preprocessing like Figure 2 As shown, firstly, the raw BeiDou GNSS observation data stream is obtained from the radio frequency link or data interface of the existing BeiDou GNSS reference station, and meteorological sensor data is collected from the newly added integrated expansion module. The integrated expansion module integrates a multi-frequency BeiDou receiver unit, a temperature, humidity, and pressure sensor, a rain gauge interface, an edge computing MCU, and a 4G / BeiDou short message dual communication module, with a total power consumption of ≤6.5W. The raw BeiDou GNSS observation data stream includes carrier phase, pseudorange, Doppler, signal-to-noise ratio, and observation timestamp. The meteorological sensor data includes air pressure, temperature, and relative humidity.

[0026] After obtaining the raw BeiDou GNSS observation data stream and the meteorological sensor data, the two are uniformly registered to the same time reference. Specifically, the observation time output by the original BeiDou GNSS receiver is unified to the BeiDou GNSS time or Beijing time service time axis, so that the BeiDou GNSS observation data and the meteorological sensor data are aligned in time to ensure the accuracy of subsequent precise single-point positioning calculation and water vapor inversion.

[0027] For existing BeiDou GNSS reference stations from different manufacturers and of different models, the integrated expansion module provides two optional data acquisition channels: Radio frequency power division multiplexing data channel and serial port / network data forwarding channel. The optimal choice should be selected based on available conditions during on-site modifications.

[0028] (II) BeiDou GNSS Water Vapor Inversion At the edge, precise single-point positioning calculations are performed on the raw BeiDou GNSS observation data stream to obtain the total zenith delay (ZTD).

[0029] As a preferred implementation, the zenith dry delay ZHD is estimated using the Saastamoinen model based on surface pressure P, station latitude φ, and station elevation h. ZHD=0.0022768·P / (1-0.00266·cos(2φ)-0.00028·h); Where P is the surface air pressure (unit: hPa), φ is the latitude of the station, and h is the elevation of the station (unit: km).

[0030] After obtaining the zenith dry delay ZHD, subtract the zenith dry delay ZHD from the zenith wet delay ZWD: ZWD = ZTD - ZHD; After obtaining the zenith wet delay (ZWD), the weighted average temperature T is calculated based on the ground temperature and relative humidity. m Weighted average temperature T m The following empirical formula can be used for calculation: T m =70.2+0.72·T s ; Among them, T s Here, T represents the ground temperature (in K). This formula is a simplified form of the Bevis formula and is applicable to mid-latitude regions. For different latitude regions, other regionalized T values ​​can also be used. m Model.

[0031] As another implementation, the weighted average temperature T m Alternatively, an empirical model incorporating humidity correction can be used to calculate: T m =a0+a1·T s +a2·e s , where e s The surface water vapor pressure (unit: hPa) is calculated from the relative humidity: e s =(RH / 100)×6.112×exp(17.67·t / (t+243.5)), where RH is relative humidity (%) and t is temperature in Celsius (°C); coefficients a0, a1, and a2 are determined by regression analysis of local radiosonde data. It should be noted that the above simplified Bevis formula is applicable to mid-latitude regions (approximately 20°N~60°N), T m The typical error is ±3~5K, corresponding to a PWV error of about 1%~2%; for low-latitude regions, a regionalized Tm model is recommended.

[0032] Get T m Then, the zenith wet delay (ZWD) is converted into atmospheric precipitable water volume (PWV): PWV=Π·ZWD Where Π is the conversion factor, Π=10 6 / [ρw·Rv·(k3 / Tm+k2′)], where ρw is the density of liquid water, Rv is the water vapor constant, and k3 and k2′ are both atmospheric refractive index constants. The dimension of k3 is K. 2 / hPa, k2′ has the dimension of K / hPa. Through the above calculation, a time-series data package of atmospheric precipitable water is generated.

[0033] Specifically, the values ​​of the above constants are: ρw = 1000 kg / m 3; Rv=461.5 J / (kg·K); k2′=22.1±2.2 K / hPa; k3=(3.739±0.012)×10 5 K 2 / hPa. Substitute T m ≈280K, we can calculate Π≈0.156, that is, PWV(mm)≈0.156×ZWD(mm).

[0034] Preferably, after generating the atmospheric precipitable water data package, the atmospheric precipitable water inversion results are compared with the atmospheric precipitable water reference values ​​from radiosondes or microwave radiometers located at existing BeiDou GNSS reference stations within a preset distance threshold. Based on the comparison results, the precise point positioning calculation parameters and the atmospheric precipitable water conversion factor are calibrated. Subsequent atmospheric precipitable water conversion is then performed using the calibrated precise point positioning calculation parameters and atmospheric precipitable water conversion factor. This reference calibration step can be performed during initial installation or periodic maintenance.

[0035] (III) Anti-interference quality control like Figure 3 As shown, anti-interference quality control is performed on the generated atmospheric precipitable water time-series data packets. This step calculates the signal quality index for each epoch in real time, identifies abnormal epochs based on the signal quality index, performs time-series repair on the atmospheric precipitable water data of abnormal epochs, and generates quality-tagged atmospheric precipitable water time-series data packets.

[0036] Specifically, the following five signal quality metrics are calculated in real time for each epoch: Valid number of satellites; Position accuracy factor (PDOP); Signal-to-noise ratio (SNR); Week jump rate; Precision single-point positioning to calculate residuals.

[0037] Based on the above signal quality indicators, abnormal epochs are identified according to the following conditions: The number of valid satellites is less than a preset threshold (e.g., less than 5). The signal-to-noise ratio decreases by more than three times the mean of the regional baseline (3σ). The cycle slip rate exceeds the preset cycle slip threshold.

[0038] The regional baseline mean is the epochal mean of the signal-to-noise ratio (SNR) over at least 72 hours of continuous clear and stable weather at this station. The standard error σ is the standard deviation of the SNR residuals within this time window. The regional baseline mean is updated every 30 days. The cycle slip rate is defined as the ratio (×100%) of the number of satellites experiencing cycle slips within a unit time window (e.g., 10 minutes) to the number of valid satellites. The preset cycle slip threshold is, for example, 20%.

[0039] If any of the above conditions are met, the current epoch will be marked as an abnormal epoch.

[0040] For epochs marked as anomalous, Empirical Mode Decomposition (EMD) or Kalman filtering is used for time series restoration. During the restoration process, the surface saturated vapor pressure from meteorological sensor data is used as an auxiliary constraint in the time series restoration calculation to improve restoration accuracy.

[0041] Specifically, the surface saturated water vapor pressure e s Based on the stated temperature, e s =6.112×exp(17.67·t / (t+243.5)) (unit: hPa), where t is the temperature in Celsius (unit: °C). One way to implement the auxiliary constraint is: using e s To calculate the upper limit of saturated atmospheric precipitable water (PWVmax), the state update result of the Kalman filter is limited to the interval [0, PWVmax]. Another implementation method is to use e... s A virtual observation equation is constructed based on the difference between the measured water vapor pressure and the actual water vapor pressure, and then incorporated into the observation update using Kalman filtering.

[0042] When the duration of continuous abnormality exceeds a preset time threshold T max (e.g., T) max When the time interval exceeds 15 minutes, the system stops pushing atmospheric precipitation data to the cloud-based AI rainfall prediction model and generates a data missing marker, which is then uploaded to the cloud. This avoids input gaps and prediction failures in the AI ​​model due to prolonged data anomalies.

[0043] Finally, the repaired epochs, the epochs with missing data markers, and the normal epochs not marked as anomalous are merged to generate a quality-labeled atmospheric precipitable water time-series data package.

[0044] (iv) Data formatting and model initialization After generating the atmospheric precipitable water time series data package, data formatting and model initialization are performed.

[0045] First, the integrated extension module is registered to a cloud-based AI computing platform. As a specific implementation, the cloud-based AI computing platform is a model training and distribution platform deployed in the cloud. After registration, a pre-trained Transformer rainfall prediction model matching the region of the existing BeiDou GNSS reference station is downloaded from the cloud-based AI computing platform. Based on the historical precipitable water time-series data and historical rainfall data of the existing BeiDou GNSS reference station, the Transformer rainfall prediction model is fine-tuned to generate an adjusted Transformer rainfall prediction model. This fine-tuning step adapts the pre-trained model to the data characteristics and local rainfall patterns of this station.

[0046] As a specific implementation method, a pre-trained Transformer rainfall prediction model is stored in the cloud-based AI computing platform. This pre-trained model is obtained by pre-training with paired atmospheric precipitable water data and rain gauge data from at least 50 stations within the region for at least 2 years. The input to the Transformer rainfall prediction model is the 15-minute atmospheric precipitable water sequence (96 time points) of the past 24 hours and the corresponding quality control labels, with an input feature dimension of at least 6. The model has a 4-layer encoder, 8 attention heads, and a hidden layer dimension of 256. The output is the rainfall level or rainfall amount for the next 1 hour, 3 hours, and 6 hours. The pre-trained Transformer rainfall prediction model is obtained by pre-training with paired atmospheric precipitable water data and rain gauge data from at least 50 stations within the region for at least 2 years; the fine-tuning learning rate is 1×10⁻⁶. -4 The batch size is 32, the iteration is 20 rounds, and the loss function is mean squared error (MSE). The fine-tuned model inference can be achieved in two ways: the fine-tuned model can be deployed at the edge for inference and the inference results can be sent back to the cloud; or the atmospheric precipitable water time-series data package can be uploaded and inferred uniformly by the cloud.

[0047] Secondly, the atmospheric precipitable water time-series data packets are aggregated into multiple time resolution levels at preset intervals. As a specific implementation, data is aggregated into 1-hour and 3-hour time resolutions at 15-minute intervals. After aggregation, data channel type and quality control tags are added to each epoch of the aggregated data, generating formatted and quality-tagged atmospheric precipitable water time-series data packets.

[0048] Quality control markers include the number of effective satellites, position accuracy factor, signal-to-noise ratio, cycle slip rate, precise single-point positioning solution residuals, and anomaly epoch markers. Data channel type markers indicate whether the current atmospheric precipitable water data originates from an RF power-division multiplexed data channel or a serial / network data forwarding channel.

[0049] Finally, the formatted and quality-labeled atmospheric precipitable water time-series data packets are input into the adjusted Transformer rainfall prediction model, and the reliability of the input data is dynamically weighted. Specifically, the Transformer rainfall prediction model assigns different reliability weights to data from different sources and of different qualities based on data channel type labels and quality control labels. For example, data from radio frequency power division multiplexing channels can be given higher weights, while data with low signal-to-noise ratios can be given lower weights, thereby improving the accuracy of rainfall prediction.

[0050] Specifically, dynamically weighting the credibility of the input data includes calculating the credibility weight w of the input data for each epoch as follows: i :w i =w0·sigmoid((SNRi -42) / 3)·sigmoid((6-PDOP i ) / 2), where SNR i Let PDOP be the signal-to-noise ratio of the i-th epoch. i is the position precision factor for the i-th epoch, and w0 is the basic weight determined according to the data channel type. The weight is 1.0 for RF power division multiplexing data channels and 0.8 for serial port / network data forwarding channels. For data with abnormal epoch markers, the weight is multiplied by 0.3 based on w0.

[0051] (V) Rainfall Forecast The time-series data of atmospheric precipitable water, labeled with quality parameters, is pushed to a cloud-based AI rainfall prediction model. This model then outputs short-term rainfall warnings for a predetermined time period, such as 1 hour, 3 hours, or 6 hours in advance. These short-term rainfall warnings are used for geological disaster monitoring and early warning.

[0052] (vi) Dual-channel BeiDou GNSS data acquisition adaptation like Figure 4 , Figure 5 As shown in this embodiment, when obtaining the raw BeiDou GNSS observation data stream from the radio frequency link or data interface of the existing BeiDou GNSS reference station, the data channel is selected according to the following method.

[0053] First, obtain the availability status of the original manufacturer's protocol for existing BeiDou GNSS reference stations and the impact level of the original displacement monitoring service.

[0054] The impact level of the original displacement monitoring service is divided into three levels: high, medium, and low, based on the data port occupancy rate, CPU load, and forwarding bandwidth margin of the original Beidou GNSS receiver. The results are determined by on-site survey and evaluation and incorporated into the upgrade plan. When the impact level of the original displacement monitoring service is high or medium, it is determined that accessing the data forwarding channel will affect the original displacement monitoring service.

[0055] Determine the availability status of the original manufacturer's agreement: If the availability status of the original protocol is that the protocol is not open or the original observation values ​​cannot be stably output, then the radio frequency power division multiplexing data channel is selected as the target data channel, and the Beidou GNSS original observation value data stream is obtained through the radio frequency power division multiplexing data channel. If the availability of the original protocol is open and it can stably output the original observation values, then the impact level of the original displacement monitoring service will continue to be determined. If the impact level of the original displacement monitoring service is determined to be that the access data forwarding channel will affect the original displacement monitoring service, then the radio frequency power division multiplexing data channel is selected as the target data channel and the Beidou GNSS raw observation data stream is obtained through the radio frequency power division multiplexing data channel. If the impact level of the original displacement monitoring service is determined to be that the access data forwarding channel does not affect the original displacement monitoring service, then the serial port / network data forwarding channel is selected as the target data channel and the BeiDou GNSS raw observation data stream is obtained through the serial port / network data forwarding channel.

[0056] It should be noted that channel selection can also consider the following indicators: data integrity requirements, RF signal margin, construction and modification costs, and power consumption increment costs. For example, if the serial / network data forwarding channel can output complete multi-frequency, multi-constellation observations, then the serial / network data forwarding channel should be preferred; if the forwarded data lacks key quality fields such as signal-to-noise ratio, cycle slip, and observation frequency, then the RF power divider / multiplexer channel should be selected. Furthermore, if the original site's antenna gain, feeder loss, and receiver carrier-to-noise ratio (C / N0) have sufficient margin, then the RF power divider / multiplexer channel can be used; if the site signal is weak, and the 3-4 dB loss introduced by the power divider will lead to a decrease in carrier-to-noise ratio and an increase in cycle slip, then the serial / network data forwarding channel should be preferred. Moreover, the serial / network forwarding channel has low construction costs and no RF loss, but it relies on the original manufacturer's protocol and port permissions; the RF power divider / multiplexer channel is highly independent and protocol-independent, but it requires the installation of a power divider, a low-noise amplifier (LNA), and RF connectors, making construction slightly more complex.

[0057] The specific implementation methods of the two data channels are described below.

[0058] 1. Radio frequency power division multiplexing data channel The specific steps for obtaining the raw BeiDou GNSS observation data stream via the radio frequency power division multiplexing data channel are as follows: The BeiDou GNSS radio frequency signal output from the original BeiDou GNSS antenna feeder of the existing BeiDou GNSS reference station is input into the broadband radio frequency power divider. The original BeiDou GNSS antenna feeder is the signal transmission feeder between the original BeiDou GNSS antenna and the original BeiDou GNSS receiver of the existing BeiDou GNSS reference station; The broadband RF power divider splits the input BeiDou GNSS RF signal into a main RF signal and a branch RF signal. The main RF signal continues to be sent to the original BeiDou GNSS receiver to ensure that the original geological disaster displacement monitoring service is not interrupted; the branch RF signal is fed into the newly added multi-frequency BeiDou receiver module after gain compensation by a low-noise amplifier (LNA). The insertion loss introduced by the broadband RF power divider is controlled within 3~4dB.

[0059] The gain compensation target of the low-noise amplifier (LNA) is: Without introducing significant noise amplification and nonlinear distortion, the carrier-to-noise ratio (C / N0) of each frequency point of the newly added multi-frequency BeiDou receiver module is restored to above the threshold for stable tracking.

[0060] As a preferred implementation, the gain compensation of the low-noise amplifier (LNA) employs a closed-loop adaptive adjustment method: The newly added multi-frequency BeiDou receiver module samples the actual signal-to-noise ratio (SNR) of each frequency point in real time, compares the actual SNR with the preset target SNR range, and calculates the SNR deviation value. When the SNR deviation value indicates that the actual SNR is lower than the lower limit of the target SNR range, the edge computing MCU generates a gain boost control signal to gradually increase the gain step of the low noise amplifier (LNA) until the actual SNR rises back to the target SNR range. As the gain step of the low-noise amplifier (LNA) is gradually increased, the edge computing MCU synchronously monitors the noise figure and signal distortion index of the LNA at each frequency point. When the noise figure exceeds the preset noise tolerance or the signal distortion index exceeds the preset distortion threshold, the edge computing MCU generates a gain back-off control signal to back the gain of the low noise amplifier (LNA) to the previous effective step, so as to maintain a dynamic balance between signal attenuation compensation and noise amplification suppression. When the actual SNR is consistently below the lower limit of the target SNR range and the gain of the low-noise amplifier (LNA) has been adjusted to the maximum allowable gain but still cannot be restored, an RF link anomaly flag is generated and a channel switching suggestion is triggered and reported to the cloud.

[0061] The newly added multi-frequency BeiDou receiver module acquires, tracks, and demodulates the gain-compensated branch radio frequency signals to generate an independent BeiDou GNSS raw observation data stream. This process does not depend on the original receiver protocol, and therefore is suitable for closed scenarios with the original manufacturer's protocol.

[0062] The input data / signals of this RF power division multiplexing channel include: the RF signal output from the original BeiDou GNSS antenna feeder; antenna gain, feeder length, and feeder loss parameters; original site coordinates, antenna height, and antenna phase center parameters; the original observation values ​​output by the newly added multi-frequency BeiDou receiving module (including carrier phase, pseudorange, Doppler, signal-to-noise ratio, and observation timestamp); and the air pressure P, temperature T, and relative humidity RH output by the temperature, humidity, and pressure sensor.

[0063] The output data of this radio frequency power division multiplexing channel includes: RINEX / binary raw observation stream generated by the newly added receiver module; PPP solution results (site coordinates, receiver clock error, ZTD); water vapor inversion results (ZWD, PWV); quality control results (effective satellite count, PDOP, signal-to-noise ratio, cycle slip rate, residual, low quality marker); and PWV time series data packets for the cloud platform.

[0064] 2. Serial port / network data forwarding channel The specific steps for obtaining the raw BeiDou GNSS observation data stream via the serial port / network data forwarding channel are as follows: First, the data output interface of the original BeiDou GNSS receiver connected to the existing BeiDou GNSS reference station is used to receive RINEX observation files, RTCM streams, or manufacturer binary observation streams output by the original BeiDou GNSS receiver, and to perform protocol parsing. Data output interfaces include RS-232, RS-485, Ethernet, TCP / UDP, or serial port servers, etc. After protocol parsing, each epoch undergoes a field integrity check to confirm whether it contains all the fields required for PWV inversion, including carrier phase, pseudorange, Doppler, signal-to-noise ratio, cycle slip identifier, and observation timestamp. If any field is incomplete, it is marked as unusable for PWV calculation or switched to the RF power division multiplexing data channel. If the field integrity check is passed, the observation time of each epoch is unified to the same time base, and the observation data from different manufacturers are converted into a unified internal format, a BeiDou GNSS raw observation data stream is generated.

[0065] The input data for this serial port / network data forwarding channel comes from the original BeiDou GNSS receiver's data interface, including: RINEX observation files; RTCM stream or manufacturer's binary raw observation stream; carrier phase, pseudorange, Doppler, signal-to-noise ratio, observation timestamp; ephemeris data, clock corrections, antenna parameters; original site coordinates, elevation, antenna height; and P, T, RH outputs from temperature, humidity, and pressure sensors.

[0066] The output data of this serial port / network data forwarding channel includes: The unified format of BeiDou GNSS raw observation data after protocol parsing; ZTD, ZWD, and PWV obtained by PPP calculation; data integrity check results; quality markers such as interruption, missing measurements, time asynchrony, and missing fields; and PWV timing and quality control data packets for cloud platforms.

[0067] (vii) End-to-Cloud Consistency Verification like Figure 6 As shown, after the quality-tagged atmospheric precipitable water time-series data package is pushed to the cloud-based AI rainfall prediction model, edge-cloud consistency verification is also performed. Edge-cloud consistency verification refers to the consistency review of the PWV calculation results between the edge device and the cloud.

[0068] Specifically, the edge device uploads the raw BeiDou GNSS observation data stream, intermediate results of precise single-point positioning calculation, calculation results of atmospheric precipitable water, and signal quality indicators to the cloud.

[0069] The cloud-based system uses the same precise single-point positioning calculation method and atmospheric precipitable water inversion method as the edge-based system, based on the raw BeiDou GNSS observation data stream, for verification, resulting in the cloud-verified atmospheric precipitable water. Both the edge and cloud-based systems employ the same core PWV calculation method: precise single-point positioning calculation based on raw BeiDou GNSS observations to obtain ZTD, combined with surface air pressure estimation of ZHD, calculating ZWD, and finally obtaining PWV through a conversion factor. The difference lies in the following: the edge-based system emphasizes real-time performance, using a modified module for local PPP calculation to quickly output PWV time series for on-site early warning, initial data screening, and local caching in case of network outages; the cloud-based system emphasizes consistency verification and quality diagnosis.

[0070] The cloud compares the atmospheric precipitable water volume (PWV) calculation results at the edge of the sliding window. edge Verify atmospheric precipitable water volume (PWV) with cloud data cloud Calculate the difference in atmospheric precipitable water ΔPWV: ΔPWV=|PWV edge -PWV cloud | A consistency anomaly alarm is triggered when the ΔPWV exceeds a preset threshold δ within K consecutive epochs, or when the proportion of anomalous epochs determined independently by either the edge or cloud end based on signal quality indicators exceeds a preset proportion threshold (e.g., 50%). Here, K is, for example, 3-5 epochs, and δ is, for example, 2-3 mm (approximately twice the uncertainty of precise single-point positioning inversion of atmospheric precipitable water). The purpose of adding this anomalous epoch proportion condition is to allow the edge and cloud ends to independently verify whether atmospheric precipitable water is abnormal, based on their respective signal quality indicators. This avoids situations where both ends are simultaneously anomalous and their changes are consistent, resulting in ΔPWV not exceeding the threshold and the alarm not being triggered.

[0071] After a consistency anomaly alarm is triggered, the cause of the anomaly can be located based on the following conditions: If the atmospheric precipitable water at the edge and the atmospheric precipitable water verified by the cloud are both abnormal, and the signal-to-noise ratio decreases, the cycle slip rate increases, and the position accuracy factor deteriorates, then it is determined to be BeiDou GNSS signal interference. If the atmospheric precipitable water content verified in the cloud is normal, but the atmospheric precipitable water content deviation at the edge continuously exceeds the preset threshold, it is determined to be an edge module failure. If timestamp misalignment, missing fields, abnormal satellite number, or incorrect observation frequency mapping occurs when data is obtained through the serial port / network data forwarding channel of the data interface, it is determined to be a protocol parsing error. If the zenith wet delay calculated at the edge is consistent with the zenith wet delay verified at the cloud, but the atmospheric precipitable water content is abnormally low and the air pressure, temperature or relative humidity changes abruptly, then the meteorological sensor is determined to be abnormal. The system identifies BeiDou GNSS signal interference, edge module failure, protocol parsing errors, and meteorological sensor anomalies as anomaly causes, and outputs corresponding maintenance recommendations based on the cause type. These recommendations include channel switching suggestions, module restart suggestions, feeder or low-noise amplifier inspection suggestions, meteorological sensor calibration suggestions, or original manufacturer protocol adaptation suggestions.

[0072] Therefore, the edge-cloud consistency verification achieves a closed-loop assessment of PWV credibility through real-time edge computing, cloud-based algorithm verification, and joint diagnosis of quality indicators. It can distinguish between external interference from BeiDou GNSS, edge hardware failures, protocol parsing errors, and meteorological sensor anomalies, thus preventing abnormal PWVs from being misjudged as precursors to heavy rainfall by AI models.

[0073] (viii) Transformation process The geological disaster monitoring method based on short-term rainfall early warning using BeiDou GNSS reference stations in this embodiment will be modified according to the following procedure during implementation: Figure 7 As shown.

[0074] S1. On-site investigation.

[0075] Check the site elevation angle (≥15°) without obstructions, the representativeness of the meteorological sensor installation environment, the redundancy capacity of the original power supply system, and the redundancy status of the original communication link. If there are tall buildings or trees obstructing the site, assess their impact on BeiDou GNSS signal reception and adjust the antenna position if necessary. The meteorological sensor should be installed in a well-ventilated, unobstructed location to ensure the representativeness of air pressure, temperature, and relative humidity measurements.

[0076] S2. Select the data channel.

[0077] Choose between the RF power division multiplexing data channel or the serial / network data forwarding channel based on whether original manufacturer agreements are available and the level of impact on the original displacement monitoring service. For specific selection methods, refer to (VI) Dual-channel BeiDou GNSS data acquisition and adaptation.

[0078] S3. Install expansion modules and expand power supply capacity.

[0079] The integrated expansion module is installed in the existing BeiDou GNSS reference station cabinet and connected to the RF feeder, data cable, and power cable. Based on the redundancy capacity of the original power supply system and the local average annual sunshine and number of consecutive rainy days, the expansion capacity of the solar panels and batteries is calculated, and the solar panels and batteries are installed according to the expansion capacity to ensure that the integrated expansion module can still work normally under continuous rainy weather.

[0080] S4. Complete PPP parameter configuration and PWV benchmark calibration.

[0081] Configure the parameters required for precise single-point positioning calculation, including antenna phase center correction, tropospheric mapping function, and Earth rotation correction. Compare the atmospheric precipitable water inversion results of this station with the atmospheric precipitable water reference values ​​of a nearby radiosonde or microwave radiometer, and calibrate the precise single-point positioning calculation parameters and the conversion factor of atmospheric precipitable water based on the comparison results.

[0082] S5. Register to the AI ​​computing platform.

[0083] The integrated extension module is registered to the cloud AI computing platform, and a pre-trained Transformer rainfall prediction model matching the area where the existing BeiDou GNSS reference station is located is downloaded from the cloud AI computing platform. The Transformer rainfall prediction model is then fine-tuned based on the historical atmospheric precipitable water time series data and historical rainfall data of this station.

[0084] S6. After verification, switch to business operation.

[0085] The upgraded system underwent joint debugging and testing to verify the proper functioning of various functional modules, including the acquisition of BeiDou GNSS raw observation data streams, meteorological sensor data collection, PWV inversion, anti-interference quality control, edge-cloud consistency verification, and rainfall prediction. After passing the tests, the system was switched to operational status.

[0086] (ix) Geological disaster monitoring based on short-term rainfall warning results Based on the short-term rainfall warning results output in step (5), geological hazards are monitored. Geological hazards include at least one of landslides, collapses, and ground subsidence.

[0087] After obtaining the short-term rainfall warning results, the rainfall amount and intensity information for the future preset time period are extracted from the short-term rainfall warning results, and the trigger risk level of the rainfall-induced factors on geological disasters is determined based on the rainfall amount and intensity information.

[0088] As a specific implementation method, the trigger risk level is determined according to the following criteria: Accumulated rainfall of less than 10 mm within a preset future period (e.g., 3 hours) is Level IV; 10-25 mm is Level III; 25-50 mm with a maximum rainfall intensity greater than 10 mm / h is Level II; and greater than 50 mm or a maximum rainfall intensity greater than 20 mm / h is Level I. The above classification criteria can be adjusted in conjunction with local critical rainfall thresholds.

[0089] Simultaneously, original displacement monitoring data is obtained from existing BeiDou GNSS reference stations. Based on this data, it is determined whether the current geological disaster deformation trend is intensifying, generating a deformation trend assessment result. The original displacement monitoring data is obtained from existing BeiDou GNSS reference stations through carrier phase differential or precise single-point positioning calculations.

[0090] When the triggered risk level exceeds the preset risk threshold and the deformation trend is determined to be intensifying, the short-term rainfall warning result and the original displacement monitoring data are input into the comprehensive geological disaster assessment module (the module includes a comprehensive geological disaster assessment model) to generate comprehensive geological disaster early warning information. This comprehensive geological disaster early warning information can be used to guide the prevention and emergency response to geological disasters.

[0091] As a specific implementation method, the criteria for determining an increased deformation trend are: the displacement rate in the past 24 hours exceeds twice the average displacement rate of the past 7 days, or the cumulative daily displacement exceeds 10 mm.

[0092] The comprehensive geological disaster assessment model outputs a comprehensive risk index R: R = 0.6·Rrainfall + 0.4·Rdeformation, where Rrainfall is the rainfall-induced risk component normalized to [0,1] according to the aforementioned trigger risk level, and Rdeformation is the deformation risk component normalized to [0,1] according to the aforementioned deformation trend judgment result. A red warning is output when R ≥ 0.8, an orange warning is output when 0.6 ≤ R < 0.8, and a yellow warning is output when 0.4 ≤ R < 0.6.

[0093] Example 2: Modification Experiment and Effect Verification To verify the low-cost, high-reliability geological disaster monitoring and early warning effect of this invention, three existing BeiDou GNSS reference stations in a geologically hazardous area in South China were selected for upgrade and transformation tests. Two of the stations had their original factory protocols closed, using radio frequency power division multiplexing (RF power division multiplexing) data channels; one station had its original factory protocols open and its original displacement monitoring services permitted, using a serial port / network data forwarding channel. The upgrades to all three stations only required the addition of an integrated expansion module (total power consumption ≤ 6.5W) within the cabinet and the completion of feeder or data interface access. No relocation or pole erection was required. The construction period for each station was 1-2 days, and the upgrade cost per station was approximately 1 / 5 to 1 / 3 of that of a newly built independent BeiDou GNSS areal rain gauge.

[0094] After the renovation was completed, statistical analysis was performed on the continuous observation data during the flood season. The results are as follows: (1) Water vapor inversion accuracy: The atmospheric precipitable water inverted by the modified station is compared with the nearby sounding data. The root mean square error RMSE ≤ 3mm and the correlation coefficient ≥ 0.9, which meets the input accuracy requirements for short-term precipitation prediction. (2) Rainfall warning performance: Based on the cloud AI rainfall prediction model outputting the short-term rainfall warning results for the next 0-6 hours, the hit rate POD ≥ 0.80, false alarm rate FAR ≤ 0.25, and TS score ≥ 0.65 for rainfall events with a rainfall of 0.1 mm / h or more, and the warning lead time for rainstorm events is 1-3 hours; (3) Accuracy of anomaly attribution: During the test, the accuracy of the end-cloud consistency verification mechanism in identifying four types of anomalies—BeiDou GNSS signal interference, edge module failure, protocol parsing error, and meteorological sensor anomaly—was no less than 90%, and the average fault location time was shortened from several hours required by traditional segment-by-segment investigation to less than 15 minutes. (4) Business continuity: During and after the transformation, the data integrity rate of the original displacement monitoring business remained above 99%, and no interruption of the original business occurred due to the transformation access during the test period.

[0095] The above test results show that the present invention can upgrade and transform existing BeiDou GNSS reference stations at low cost, obtain atmospheric precipitable water that meets the input accuracy requirements for short-term precipitation warnings while ensuring the continuity of the original displacement monitoring services, and achieve rapid and accurate location of the cause of anomalies through end-cloud consistency verification.

[0096] The embodiments of the present invention achieve the following technical effects: This invention, through the connection of a broadband radio frequency power divider across the feeder side of existing BeiDou GNSS reference stations and the use of a low-noise amplifier for gain compensation, enables newly added multi-frequency BeiDou receiving modules to independently generate raw observation values. This achieves non-intrusive acquisition of data required for water vapor inversion without interrupting existing geological disaster displacement monitoring services or relying on original manufacturer protocols. The radio frequency power divider multiplexing channel complements the serial / network data forwarding channel—radio frequency power divider is used when the protocol is closed, and serial forwarding is used when the protocol is open and services permit. Both methods share the same subsequent processing flow, resolving the access difficulties caused by the mixed manufacturer models and inconsistent interface protocols of existing stations.

[0097] At the data processing level, the total zenith delay is obtained through precise single-point positioning calculation. Then, the zenith wet delay is separated by combining ground pressure, temperature and relative humidity data and inverted into atmospheric precipitable water. Subsequently, the time series repair method of empirical mode decomposition or Kalman filtering is used, along with ground saturated water vapor pressure as an auxiliary constraint. This solves the problem of abnormal jumps in PWV time series data when BeiDou GNSS signals are interfered with, thus controlling the data quality entering the rainfall prediction model.

[0098] By deploying a consistency verification mechanism at the edge for real-time calculation and using the same algorithm for verification in the cloud, and combining multi-dimensional quality indicators such as the number of effective satellites, position accuracy factor, signal-to-noise ratio, cycle slip rate, and calculation residuals, abnormal PWV deviations can be attributed to BeiDou GNSS signal interference, edge module failures, protocol parsing errors, or meteorological sensor anomalies. This edge-cloud collaborative verification approach can quickly provide fault type identification and maintenance suggestions when PWV anomalies occur in the field, reducing the time spent troubleshooting equipment, links, and the environment one by one.

[0099] Furthermore, the upgrade of existing BeiDou GNSS reference stations only requires adding an integrated expansion module (power consumption ≤ 6.5W, solar-powered) to the cabinet and completing the connection of the feeder or serial port. A single station can cover a surrounding area of ​​at least 3 kilometers. Compared with building a new independent BeiDou GNSS areal rain gauge, there is no need to relocate and erect poles, add antennas and receivers, or independently deploy power supply and communication. The hardware cost and construction period are significantly reduced (for example, in Example 2, the upgrade cost is about 1 / 5 to 1 / 3 of that of building a new independent BeiDou GNSS areal rain gauge, and the construction period is shortened from several weeks to 1 to 2 days). By linking the short-term rainfall warning results with the original displacement monitoring data of existing stations, comprehensive geological disaster warning information is generated when the risk level triggered by rainfall exceeds the threshold and the deformation trend intensifies. This realizes the improvement of the ability from simple areal rainfall prediction to comprehensive geological disaster assessment, making it feasible in engineering to carry out large-scale geological disaster monitoring and early warning based on the existing geological disaster monitoring network and CORS station network.

[0100] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0101] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A geological disaster monitoring method based on short-term rainfall early warning using BeiDou GNSS reference stations, characterized in that, include: The BeiDou GNSS raw observation data stream is obtained from the radio frequency link or data interface of the existing BeiDou GNSS reference station, and meteorological sensor data is collected from the added integrated extension module. The BeiDou GNSS raw observation data stream and the meteorological sensor data are uniformly registered to the same time reference. The BeiDou GNSS raw observation data stream includes carrier phase, pseudorange, Doppler, signal-to-noise ratio and observation timestamp, and the meteorological sensor data includes air pressure, temperature and relative humidity. At the edge end, precise single-point positioning calculation is performed on the raw BeiDou GNSS observation data stream to obtain the total zenith delay. The zenith dry delay is estimated based on the air pressure and the station coordinates of the existing BeiDou GNSS reference stations. The zenith wet delay is obtained by subtracting the zenith dry delay from the total zenith delay. The weighted average temperature is calculated based on the temperature and the relative humidity. The zenith wet delay is converted into atmospheric precipitable water content, and atmospheric precipitable water content time-series data packets are generated. Anti-interference quality control is performed on the atmospheric precipitable water time series data packet. The signal quality index of each epoch is calculated in real time. Abnormal epochs are identified based on the signal quality index, and the atmospheric precipitable water in the abnormal epochs is time-series repaired to generate atmospheric precipitable water time series data packets with quality tags. The time-series data package of atmospheric precipitable water, which has been marked with quality, is pushed to the cloud-based AI rainfall prediction model. The cloud-based AI rainfall prediction model outputs short-term rainfall warning results for a preset period in the future, and determines the trigger risk level of geological disasters based on the short-term rainfall warning results, so as to monitor geological disasters.

2. The geological disaster monitoring method as described in claim 1, characterized in that, Obtain raw BeiDou GNSS observation data streams from the radio frequency links or data interfaces of existing BeiDou GNSS reference stations, including: Obtain the availability status of the original manufacturer's protocol and the impact level of the original displacement monitoring service for the existing BeiDou GNSS reference stations; Determine the availability status of the original equipment manufacturer (OEM) agreement; If the availability status of the original protocol is that the protocol is not open or the original observation value cannot be stably output, then the radio frequency power division multiplexing data channel is selected as the target data channel, and the Beidou GNSS original observation value data stream is obtained through the radio frequency power division multiplexing data channel. If the availability status of the original protocol is that the protocol is open and can stably output the original observation values, then the impact level of the original displacement monitoring service is determined. If the determination result of the impact level of the original displacement monitoring service is that the access data forwarding channel will affect the original displacement monitoring service, then the radio frequency power division multiplexing data channel is selected as the target data channel and the Beidou GNSS raw observation data stream is obtained through the radio frequency power division multiplexing data channel. If the determination result of the impact level of the original displacement monitoring service is that the access data forwarding channel does not affect the original displacement monitoring service, then the serial port / network data forwarding channel is selected as the target data channel and the Beidou GNSS raw observation data stream is obtained through the serial port / network data forwarding channel.

3. The geological disaster monitoring method as described in claim 2, characterized in that, The BeiDou GNSS raw observation data stream is obtained via the radio frequency power division multiplexing data channel, including: The BeiDou GNSS radio frequency signal output from the original BeiDou GNSS antenna feeder of the existing BeiDou GNSS reference station is input into a broadband radio frequency power divider. The broadband radio frequency power divider divides the signal into a main radio frequency signal and a branch radio frequency signal. The original BeiDou GNSS antenna feeder is the signal transmission feeder between the original BeiDou GNSS antenna and the original BeiDou GNSS receiver of the existing BeiDou GNSS reference station. The main radio frequency signal is continued to be sent to the original Beidou GNSS receiver, and the branch radio frequency signal is input into the newly added multi-frequency Beidou receiver module after gain compensation by a low-noise amplifier. The newly added multi-frequency BeiDou receiving module captures, tracks, and demodulates the branch radio frequency signal after gain compensation, generating the BeiDou GNSS raw observation data stream.

4. The geological disaster monitoring method as described in claim 3, characterized in that, The branch radio frequency signal is subjected to gain compensation via a low-noise amplifier, including: The actual signal-to-noise ratio of each BeiDou GNSS frequency point is sampled in real time, and the actual signal-to-noise ratio is compared with the preset target signal-to-noise ratio range; When the actual signal-to-noise ratio is lower than the lower limit of the target signal-to-noise ratio range, the gain step of the low-noise amplifier is gradually increased until the actual signal-to-noise ratio rises back to the target signal-to-noise ratio range; As the gain step of the low-noise amplifier is gradually increased, the edge computing MCU synchronously monitors the noise figure and the signal distortion index at each frequency point. When the noise figure exceeds the preset noise tolerance or the signal distortion index exceeds the preset distortion threshold, the gain of the low-noise amplifier is reverted to the step value before the current gain increase. When the actual signal-to-noise ratio is consistently lower than the lower limit of the target signal-to-noise ratio range and the low-noise amplifier's gain has been adjusted to the maximum allowable gain but still cannot recover, an RF link anomaly flag is generated and a channel switching suggestion is triggered and reported to the cloud.

5. The geological disaster monitoring method as described in claim 2, characterized in that, The BeiDou GNSS raw observation data stream is obtained via the serial port / network data forwarding channel, including: The original BeiDou GNSS receiver connected to the existing BeiDou GNSS reference station receives the RINEX observation file, RTCM stream, or manufacturer binary observation stream output by the original BeiDou GNSS receiver through the data output interface, and performs protocol parsing. Perform a field integrity check on each epoch after protocol parsing to confirm whether each epoch contains carrier phase, pseudorange, Doppler, signal-to-noise ratio, cycle slip identifier and observation timestamp; If the field integrity check is passed, the observation time of each epoch is unified to the same time base, and the observation data from different manufacturers are converted into a unified internal format, the BeiDou GNSS raw observation data stream is generated.

6. The geological disaster monitoring method as described in claim 1, characterized in that, Anti-interference quality control is performed on the atmospheric precipitable water time-series data packet. The signal quality index for each epoch is calculated in real time. Abnormal epochs are identified based on the signal quality index, and the atmospheric precipitable water data in the abnormal epochs is time-series repaired to generate a quality-tagged atmospheric precipitable water time-series data packet, including: The effective number of satellites, position accuracy factor, signal-to-noise ratio, cycle slip rate, and precision single-point positioning solution residuals are calculated in real time for each epoch and used as the signal quality indicators. When the number of effective satellites is less than a preset threshold, the signal-to-noise ratio drops by more than three times the mean of the regional baseline, or the cycle slip rate exceeds a preset threshold, the current epoch is marked as an abnormal epoch. Empirical mode decomposition or Kalman filtering is used, and the surface saturated water vapor pressure in the meteorological sensor data is used as an auxiliary constraint to perform time-series repair on the atmospheric precipitable water in the anomalous epoch, generating a repaired epoch. When the duration of continuous anomalies exceeds a preset time threshold, the push of atmospheric precipitable water data to the cloud-based AI rainfall prediction model is stopped, and a data missing marker is generated and uploaded to the cloud. The repaired epochs, the epochs that generated the data missing markers, and the normal epochs that were not marked as anomalous are merged to generate the quality-marked atmospheric precipitable water time-series data package.

7. The geological disaster monitoring method according to any one of claims 1 to 6, characterized in that, After generating the atmospheric precipitable water time-series data package, the following is also included: Register the integrated extension module to the cloud AI computing platform, download the pre-trained Transformer rainfall prediction model that matches the area where the existing BeiDou GNSS reference station is located from the cloud AI computing platform, and adjust the Transformer rainfall prediction model based on the historical atmospheric precipitable water time series data and historical rainfall data of the existing BeiDou GNSS reference station to generate the adjusted Transformer rainfall prediction model. The atmospheric precipitable water data packets are aggregated into multiple time resolution levels at preset intervals, and data channel type markers and quality control markers are added to each epoch after aggregation to generate formatted and quality-marked atmospheric precipitable water data packets. The quality control markers include the number of effective satellites, position accuracy factor, signal-to-noise ratio, cycle slip rate, precise single-point positioning solution residuals, and abnormal epoch markers. The data channel type markers are used to identify whether the current atmospheric precipitable water data comes from a radio frequency power division multiplexing data channel or a serial port / network data forwarding channel. The formatted and quality-tagged atmospheric precipitable water time series data package is input into the adjusted Transformer rainfall prediction model, and the reliability of the input data is dynamically weighted.

8. The geological disaster monitoring method according to any one of claims 1 to 6, characterized in that, After pushing the quality-tagged atmospheric precipitable water time-series data package to the cloud-based AI rainfall prediction model, the process also includes: The edge device uploads the BeiDou GNSS raw observation data stream, the intermediate results of the precise single-point positioning calculation, the calculation results of atmospheric precipitable water, and the signal quality indicators to the cloud. The cloud-based system uses the same precise single-point positioning solution method and the same atmospheric precipitable water inversion method as the edge end to verify the cloud-based verified atmospheric precipitable water based on the BeiDou GNSS raw observation data stream. The cloud compares the calculated atmospheric precipitable water content at the edge with the cloud's verified atmospheric precipitable water content within a sliding window, and calculates the difference in atmospheric precipitable water content. A consistency anomaly alarm is triggered when the difference in atmospheric precipitable water exceeds a preset threshold within K consecutive epochs, or when the proportion of abnormal epochs determined independently by either end based on the signal quality index exceeds a preset proportion threshold. If, after the consistency anomaly alarm is triggered, the atmospheric precipitable water at the edge end and the atmospheric precipitable water verified in the cloud are both abnormal, and the signal-to-noise ratio decreases, the cycle slip rate increases, and the position accuracy factor deteriorates, then it is determined to be BeiDou GNSS signal interference. If the cloud-based verification of atmospheric precipitable water is normal, but the deviation of atmospheric precipitable water at the edge continuously exceeds the preset threshold, it is determined to be an edge module failure. If timestamp misalignment, missing fields, abnormal satellite number, or incorrect observation frequency mapping occurs when data is obtained through the serial port / network data forwarding channel of the data interface, it is determined to be a protocol parsing error. If the zenith wet delay calculated at the edge is consistent with the zenith wet delay verified at the cloud, but the atmospheric precipitable water deviation is abnormal and the air pressure, temperature, or relative humidity changes abruptly, then the meteorological sensor is determined to be abnormal. The BeiDou GNSS signal interference, edge module failure, protocol parsing error, and meteorological sensor anomaly are identified as anomaly cause types. Based on the anomaly cause type, corresponding operation and maintenance suggestions are output. These suggestions include channel switching suggestions, module restart suggestions, feeder or low-noise amplifier inspection suggestions, meteorological sensor calibration suggestions, or original manufacturer protocol adaptation suggestions.

9. The geological disaster monitoring method as described in claim 1, characterized in that, Based on the aforementioned short-term rainfall warning results, geological disaster monitoring is conducted, including: The geological hazards include at least one of landslides, collapses, and ground subsidence; After obtaining the short-term rainfall warning result, the rainfall amount and rainfall intensity information for a future preset period are extracted from the short-term rainfall warning result, and the trigger risk level of the rainfall-induced factors on geological disasters is determined based on the rainfall amount and rainfall intensity information; Original displacement monitoring data is obtained from the existing BeiDou GNSS reference stations. Based on the original displacement monitoring data, it is determined whether the current geological disaster deformation trend is intensifying, and a deformation trend determination result is generated. When the triggered risk level exceeds the preset risk threshold and the deformation trend determination result is aggravated, the short-term rainfall warning result and the original displacement monitoring data are input into the comprehensive geological disaster assessment model to generate comprehensive geological disaster warning information.

10. A geological disaster monitoring device based on short-term rainfall early warning using a BeiDou GNSS reference station, used to execute the geological disaster monitoring method according to any one of claims 1 to 9, characterized in that, include: The original BeiDou GNSS antenna and the original BeiDou GNSS receiver of the existing BeiDou GNSS reference station, wherein the original BeiDou GNSS antenna is used to receive BeiDou GNSS radio frequency signals, and the original BeiDou GNSS receiver is used to perform the original displacement monitoring service; A broadband radio frequency power divider, whose input is connected to the original BeiDou GNSS antenna, and whose output is divided into a main path and a branch path. The main path is connected to the original BeiDou GNSS receiver, and the branch path is connected to a low-noise amplifier. Low-noise amplifiers are used to compensate for the gain of branch radio frequency signals; The multi-frequency BeiDou receiver module is connected to the output of the low-noise amplifier and is used to capture, track and demodulate the branch radio frequency signal after gain compensation to generate a BeiDou GNSS raw observation data stream. The integrated expansion module integrates the multi-frequency BeiDou receiver module, temperature, humidity and pressure sensor, rain gauge interface, edge computing MCU and 4G / BeiDou short message dual communication module. It is used to collect meteorological sensor data and perform precise single-point positioning calculation, atmospheric precipitable water inversion and anti-interference quality control at the edge end, and generate quality-tagged atmospheric precipitable water time-series data packets. It is also used to obtain the original displacement monitoring data from the existing BeiDou GNSS reference station. The original displacement monitoring data is obtained by the existing BeiDou GNSS reference station through carrier phase differential or precise single-point positioning calculation. A power management module, including a solar panel and a battery expansion kit, is used to power the integrated expansion module; The rainfall prediction module is used to push the quality-tagged atmospheric precipitable water time series data package to the cloud AI rainfall prediction model, and output the short-term rainfall warning results for the future preset period through the cloud AI rainfall prediction model. The short-term rainfall warning results are used for geological disaster monitoring and early warning. The comprehensive geological disaster assessment module is used to generate comprehensive geological disaster early warning information when the risk level of the rainfall trigger indicated by the short-term rainfall warning result exceeds a preset risk threshold and the geological disaster deformation trend determined according to the original displacement monitoring data intensifies.