Vehicle-mounted positioning method, device and equipment based on Beidou satellite signal and medium

Abstract

The invention provides a vehicle-mounted positioning method and device based on a Beidou satellite signal, equipment and a medium. The method comprises the following steps: acquiring the Beidou satellite signal; if a base station signal is received, entering a base station positioning mode, processing a Beidou satellite signal according to a single-frequency real-time dynamic difference technology to obtain satellite positioning data, inputting the satellite positioning data, inertial navigation data, mileage data and orientation data into a tight combination positioning model for fusion processing, and outputting fused positioning data; high-frequency errors and low-frequency errors generated in the positioning process are obtained, and the fused positioning data are corrected according to the high-frequency errors and the low-frequency errors to generate a vehicle positioning result. According to the method, the accident judgment efficiency is improved in the financial field, the UBI insurance charging is optimized, the vehicle asset safety is guaranteed, the emergency rescue response time is shortened in the medical health old-age care field, the old-age monitoring reliability is enhanced, vehicle-mounted medical data accurate interaction is assisted, and the advantages of autonomous controllability and low-power-consumption expansion are achieved.

Description

Technical Field

[0001] This application relates to the field of machine learning technology, and in particular to a vehicle positioning method, device, equipment and medium based on BeiDou satellite signals. Background Technology

[0002] With the development of intelligent transportation, autonomous driving, and IoT technologies, the demand for vehicle positioning systems in fields such as finance, healthcare, and elderly care is becoming increasingly urgent. In the financial sector, vehicle positioning accuracy directly affects accident liability determination, stolen vehicle tracking, and UBI (usage-based insurance) billing. Traditional positioning systems suffer from poor signal stability in complex environments (such as the "canyon effect" between tall buildings in cities or tunnel obstructions), resulting in positioning errors reaching the meter level, easily leading to claims disputes. In the healthcare and elderly care sectors, for vehicle monitoring and emergency rescue scenarios targeting the elderly or special patients, positioning systems need to continuously provide high-precision location and heading information under complex road conditions. However, existing technologies show significant accumulation of inertial navigation errors when signals are lost for extended periods (such as in tunnels), failing to meet the real-time, accurate positioning requirements of emergency rescue. Current vehicle positioning technologies in the industry mainly suffer from the following shortcomings: 1. Insufficient adaptability to complex environments: Relying on GPS or BeiDou / GPS dual-mode systems, the positioning accuracy drops sharply to the meter level in scenarios such as urban canyons and tunnels due to multipath effects and obstruction. This cannot meet the requirements of financial insurance for high-precision reconstruction of accident scenes and medical rescue for real-time location tracking.

[0003] 2. Low efficiency of multi-source fusion: Traditional compact combination models do not effectively integrate dual-antenna directional data, resulting in heading drift under the viaduct. Furthermore, they do not distinguish between high and low frequency error characteristics, resulting in low correction efficiency and an inability to continuously output stable positioning results during dynamic driving.

[0004] 3. Poor scalability of base station services: High server load and large response latency under massive concurrency, making it difficult to support the needs of large-scale real-time vehicle monitoring in the insurance industry and synchronous positioning of multiple devices in elderly care monitoring systems.

[0005] Therefore, a method is urgently needed to solve at least one of the above problems. Summary of the Invention

[0006] This application provides a vehicle positioning method, device, equipment, and medium based on BeiDou satellite signals, aiming to solve the shortcomings of current vehicle positioning technologies in the industry, such as insufficient adaptability to complex environments, low efficiency of multi-source fusion, and limited scalability of reference station services.

[0007] Firstly, this application provides a vehicle-mounted positioning method based on BeiDou satellite signals, including: Acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition; Acquire inertial navigation data collected by the vehicle's inertial navigation equipment, mileage data collected by the wheel odometer, and orientation data collected by the dual-antenna module; If a base station signal is received, the system enters the base station positioning mode. The BeiDou satellite signal is processed according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. The satellite positioning data, inertial navigation data, mileage data and orientation data are input into a preset compact combination positioning model for fusion processing, and the fused positioning data is output. The high-frequency and low-frequency errors generated during the positioning process are acquired, and the fused positioning data is corrected based on the high-frequency and low-frequency errors. The vehicle positioning result is then generated based on the corrected fused positioning data.

[0008] In some embodiments, the BeiDou satellite signal includes BeiDou-3 new frequency point signals; the step of performing multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signal to enhance weak signal acquisition includes: dividing the BeiDou-3 new frequency point signals into frequency bands according to the corresponding BeiDou-3 new frequency points, accumulating the energy of the BeiDou-3 new frequency point signals in each frequency band, and dynamically adjusting the accumulation time according to the signal strength to obtain the accumulation result; and weighting and merging the accumulation results of each frequency band to obtain the enhanced BeiDou satellite signal.

[0009] In some embodiments, before acquiring the inertial navigation data collected by the inertial navigation device mounted on the vehicle, the mileage data collected by the wheel odometer, and the orientation data collected by the dual-antenna module, the method further includes: suppressing electromagnetic interference of the BeiDou satellite signal through a preset adaptive anti-narrowband interference circuit; identifying the frequency range of narrowband interference by real-time monitoring of the signal spectrum; dynamically generating a notch filter corresponding to the interference frequency; and filtering the BeiDou satellite signal through the notch filter to remove narrowband interference components.

[0010] In some embodiments, before the satellite positioning data, inertial navigation data, mileage data, and orientation data are input into a preset compactly combined positioning model for fusion processing, the method further includes: if no base station signal is received, entering a base station-free positioning mode to process the BeiDou satellite signal using single-frequency precise single-point positioning technology to obtain satellite positioning data; and inputting the satellite positioning data, inertial navigation data, mileage data, and orientation data into the compactly combined positioning model.

[0011] In some embodiments, the step of inputting satellite positioning data, inertial navigation data, odometer data, and orientation data into a preset compactly combined positioning model for fusion processing and outputting fused positioning data includes: establishing a state-space model of satellite positioning data, inertial navigation data, odometer data, and orientation data in the compactly combined positioning model; obtaining the velocity and heading corresponding to the inertial navigation data; predicting the vehicle position based on the inertial navigation data, using the velocity and heading as state variables, and using the position information corresponding to the satellite positioning data and the displacement information corresponding to the odometer data as observation variables; iteratively updating the state variables using a Kalman filter algorithm; and introducing dual-antenna orientation data to correct and constrain the heading angle, and outputting the fused positioning data.

[0012] In some embodiments, acquiring the high-frequency and low-frequency errors generated during the positioning process includes: classifying the errors during the positioning process according to their time characteristics; classifying receiver clock error and signal acquisition noise as the high-frequency errors; estimating the high-frequency errors periodically using real-time satellite signal observations; classifying ionospheric delay and tropospheric delay as the low-frequency errors; and predicting the trend of the low-frequency errors based on historical observation data from the reference station network and real-time meteorological data.

[0013] In some embodiments, the step of correcting the fused positioning data based on the high-frequency and low-frequency errors to generate a vehicle positioning result based on the corrected fused positioning data includes: applying a real-time Kalman filter algorithm to filter and correct the high-frequency errors, compensating them into the position, velocity, and heading parameters corresponding to the fused positioning data; and for the low-frequency errors, by receiving low-frequency error correction information for regular grid points broadcast by the reference station network, matching the correction values ​​of neighboring grid points according to the vehicle's real-time position, and weighting and correcting the ionospheric and tropospheric delay errors in the fused positioning data to generate the vehicle positioning result containing position coordinates and heading angle.

[0014] Secondly, this application provides a vehicle-mounted positioning device based on BeiDou satellite signals, comprising: The signal acquisition unit is used to acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition. The data acquisition unit is used to acquire inertial navigation data collected by the inertial navigation device mounted on the vehicle, mileage data collected by the wheel odometer, and orientation data collected by the dual-antenna module. The fusion output unit is used to enter the base station positioning mode when a base station signal is received. It processes the BeiDou satellite signal according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. It inputs the satellite positioning data, inertial navigation data, mileage data and orientation data into a preset compact combination positioning model for fusion processing and outputs fused positioning data. The result generation unit is used to acquire high-frequency and low-frequency errors generated during the positioning process, correct the fused positioning data based on the high-frequency and low-frequency errors, and generate vehicle positioning results based on the corrected fused positioning data.

[0015] Thirdly, this application also provides a computer device, comprising: Memory and processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the steps of the vehicle positioning method based on BeiDou satellite signals as described in the first aspect above.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the vehicle positioning method based on BeiDou satellite signals as described in the first aspect above.

[0017] This application provides a vehicle-mounted positioning method, device, equipment, and medium based on BeiDou satellite signals. The aim is to improve signal quality in complex environments by acquiring signals containing new BeiDou-3 frequency points, enhancing weak signal acquisition capabilities through multi-frequency incoherent accumulation, and suppressing electromagnetic interference using an adaptive anti-narrowband interference circuit. By collecting inertial navigation data, wheel mileage data, and dual-antenna directional data, and depending on whether a base station signal is received, single-frequency real-time dynamic differential (RTK) or single-frequency precise point positioning (PPP) techniques are used to process the BeiDou signals. A pre-set compact combination positioning model fuses multi-source data, and dual-antenna directional constraints are introduced to address heading drift issues under viaducts. High-frequency errors (receiver clock bias) are estimated in real-time through filtering, and low-frequency errors (ionospheric / tropospheric delay) are predicted using weighted summaries and corrected via a base station network broadcast. Virtual gridded data broadcasting technology reduces server load, ultimately generating a high-precision positioning result.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1This is a schematic flowchart illustrating the steps of a vehicle positioning method based on BeiDou satellite signals provided in an embodiment of this application; Figure 2 This is a schematic flowchart illustrating the steps of a BeiDou satellite signal enhancement method according to an embodiment of this application; Figure 3 This is a schematic flowchart illustrating the steps of another vehicle positioning method based on BeiDou satellite signals provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a vehicle-mounted positioning device based on BeiDou satellite signals provided in one embodiment of this application; Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0021] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0024] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and the terms "first" and "second" are not necessarily different.

[0025] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0026] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0027] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0028] With the development of intelligent transportation, autonomous driving, and IoT technologies, the demand for vehicle positioning systems in fields such as finance and insurance, healthcare, and elderly care is becoming increasingly urgent. In the finance and insurance sector, vehicle positioning accuracy directly affects accident liability determination, stolen vehicle tracking, and UBI (usage-based insurance) billing. Traditional positioning systems suffer from poor signal stability in complex environments (such as the "canyon effect" between tall buildings in cities or tunnel obstructions), resulting in positioning errors reaching the meter level, easily leading to claims disputes. In the healthcare and elderly care sector, for vehicle monitoring and emergency rescue scenarios targeting the elderly or special patients, positioning systems need to continuously provide high-precision location and heading information under complex road conditions. However, existing technologies show significant accumulation of inertial navigation errors when signals are lost for extended periods (such as in tunnels), failing to meet the real-time, accurate positioning requirements of emergency rescue. Current vehicle positioning technologies in the industry mainly suffer from the following shortcomings: 1. Insufficient adaptability to complex environments: Relying on GPS or BeiDou / GPS dual-mode systems, the positioning accuracy drops sharply to the meter level in scenarios such as urban canyons and tunnels due to multipath effects and obstruction. This cannot meet the requirements of financial insurance for high-precision reconstruction of accident scenes and medical rescue for real-time location tracking.

[0029] 2. Low efficiency of multi-source fusion: Traditional compact combination models do not effectively integrate dual-antenna directional data, resulting in heading drift under the viaduct. Furthermore, they do not distinguish between high and low frequency error characteristics, resulting in low correction efficiency and an inability to continuously output stable positioning results during dynamic driving.

[0030] 3. Poor scalability of base station services: High server load and large response latency under massive concurrency, making it difficult to support the needs of large-scale real-time vehicle monitoring in the insurance industry and synchronous positioning of multiple devices in elderly care monitoring systems.

[0031] To address the aforementioned issues, this invention provides a vehicle-mounted positioning method based on BeiDou satellite signals. Through BeiDou signal enhancement processing, multi-source data tight combination fusion, and high- and low-frequency error diversion correction, it significantly improves positioning accuracy and reliability in complex environments, effectively solving pain points in the fields of finance, insurance, medical care, and elderly care.

[0032] Please see Figure 1 , Figure 1This is a schematic flowchart illustrating a vehicle-mounted positioning method based on BeiDou satellite signals according to an embodiment of this application. This vehicle-mounted positioning method based on BeiDou satellite signals can be implemented using computer equipment, which can be deployed on a single server or a server cluster. It can also be deployed on handheld terminals, laptops, wearable devices, or robots, etc.

[0033] It should be noted that the acquisition of any information mentioned in the provided methods is in compliance with relevant regulations and is carried out with the user's consent, and will not infringe on the user's privacy or violate relevant laws and regulations.

[0034] like Figure 1 As shown, the provided vehicle positioning method based on BeiDou satellite signals includes steps S101 to S104. Details are as follows: Step S101. Acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition.

[0035] Specifically, multi-frequency incoherent accumulation technology enhances the weak signal acquisition capability of BeiDou satellites, addressing the problem of insufficient signal strength in obstructed scenarios such as urban canyons and tunnels. Specifically, incoherent energy accumulation of carrier phase / pseudorange signals from multiple frequency points on the same satellite using BeiDou B1I / B2I / B3I multi-frequency signals improves the signal-to-noise ratio (SNR), achieving effective acquisition of ultra-weak signals below -160dBm.

[0036] Multi-frequency signal separation and filtering involves receiving BeiDou tri-frequency signals (1561.098MHz, 1268.52MHz, and 1207.14MHz), separating the signals of each frequency band using a bandpass filter, and eliminating invalid frequency bands with low signal-to-noise ratios (SNR<25dBHz).

[0037] In financial scenarios, the B1I band (high update rate) is prioritized for real-time accident location, while the B2I band (high precision) is used for post-accident trajectory reconstruction. In medical scenarios, full-band reception is used, with a focus on enhancing the B3I band (strong anti-interference capability) signal in tunnels to ensure the continuity of emergency rescue signals.

[0038] The noncoherent accumulation algorithm accumulates the energy of signals from different frequency bands of the same satellite. The expressions corresponding to BeiDou satellite signals include: ; Where Ai and Bi are the orthogonal carrier amplitudes of each frequency band, Sacc is the accumulated BeiDou satellite signal, n is the number of frequency bands, and the number of accumulations is dynamically adjusted according to the scenario (e.g., a fixed 8 accumulations in the financial scenario, and a fast 4 accumulations in the medical scenario due to the need for real-time processing). By introducing a sliding window mechanism, the accumulation window is updated every 50ms to avoid phase ambiguity caused by signal Doppler frequency shift.

[0039] In financial scenarios, such as at accident-prone urban intersections, enhanced signal capture can accurately record the trajectory 10 seconds before a collision, solving the problem of location point jumps caused by the "canyon effect" and providing continuous trajectory evidence for liability determination.

[0040] In medical scenarios, such as when driving in tunnels, enhancing the signal to maintain effective tracking by at least three satellites provides basic positioning input for the vehicle monitoring system, avoiding rescue delays caused by signal loss.

[0041] Step S102. Acquire inertial navigation data collected by the inertial navigation device mounted on the vehicle, mileage data collected by the wheel odometer, and orientation data collected by the dual antenna module.

[0042] Specifically, a multi-source heterogeneous sensor fusion system is constructed by real-time acquisition of data from the inertial navigation system (IMU), wheel odometer, and dual-antenna orientation module. The IMU provides acceleration and angular velocity (100Hz high-frequency sampling), the odometer calculates displacement through wheel speed pulses (resolution 0.1 m / pulse), and the dual-antenna module obtains the heading angle (accuracy ±0.1°) through carrier phase differential.

[0043] The IMU is installed at the vehicle's center of gravity, using MEMS devices (zero-bias stability <5° / h). It is synchronized with the BeiDou receiver via a hardware clock module with a time deviation of <1μs. The distance between the two antennas is ≥1.2 meters (baseline length), and they are deployed at the front and rear ends of the vehicle roof respectively. The heading angle is calculated using the BeiDou dual-frequency carrier phase difference: θ=arctan2(Δy,Δx), where Δx and Δy are the baseline vectors calculated by the phase difference between the two antennas.

[0044] The odometer pulse signal is filtered by Kalman filter to remove wheel slip noise. In the financial scenario, the vehicle CAN bus is used for real-time wheel speed correction (error <0.5%). In the medical scenario, a barometric pressure sensor is added to assist in slope compensation. Dual-antenna directional data is used in the financial scenario to determine the attitude of the accident vehicle (such as the rollover angle) and in the medical scenario to maintain the course continuity during emergency rescue (avoiding course drift in tunnels).

[0045] In financial scenarios, by focusing on calibrating dual-antenna directional data, the collision direction can be determined by detecting sudden changes in heading angle (threshold ±15° / s) when an accident occurs, and the collision point coordinates can be accurately reconstructed by combining the odometer displacement (error <0.5 meters).

[0046] In medical settings, the IMU sampling rate is increased to 200Hz to monitor vehicle acceleration and deceleration changes in real time. Combined with odometer data, it can construct inertial tracks when signals are lost (such as maintaining effective positioning for 30 seconds in a tunnel), providing continuous location information for emergency rescue.

[0047] Step S103. If a base station signal is received, enter the base station positioning mode. Process the BeiDou satellite signal according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. Input the satellite positioning data, inertial navigation data, mileage data and orientation data into the preset compact combination positioning model for fusion processing and output fused positioning data.

[0048] Specifically, centimeter-level satellite positioning data is acquired based on single-frequency real-time dynamic differential (RTK) technology. A state equation is constructed by fusing IMU, odometer, and dual-antenna data through a compact combination model. ; The state vector x contains 15 parameters including position, velocity, attitude, and IMU error, and the measurement equation z integrates satellite pseudorange / carrier phase, odometer displacement, and dual-antenna heading angle.

[0049] Single-frequency RTK processing receives RTCM3.3 differential data (1Hz update rate) broadcast from the base station and calculates floating-point solutions using a single-frequency integer ambiguity fixing algorithm (L1 band). In financial scenarios, a "three consecutive fixed solution verifications" mechanism is employed (to avoid random errors), while in medical scenarios, "fast ambiguity decomposition" (time <2 seconds) is used. Heading state correction is achieved by fusing dual-antenna directional data and using the heading angle calculated from the dual antennas as measurement input to correct IMU attitude drift (especially the heading error problem of traditional compact combination models under viaducts).

[0050] In financial scenarios, by increasing accident characteristic parameters (such as the emergency braking acceleration threshold of 10 m / s²), 2 The model parameters are adaptively adjusted to trigger the collision, which increases the noise covariance of the Kalman filter process 5 seconds before the collision and enhances the accuracy of trajectory fitting. In medical settings, a signal loss pre-detection mechanism is designed (triggered when the number of satellites is less than 3) to switch to the inertial / odometry dominant mode in advance, maintaining the continuity of positioning output (position error growth rate < 0.1 m / s).

[0051] In financial scenarios and in UBI billing, a compact combination model is used to output high-precision trajectories in real time (longitude / latitude error <0.5 meters, heading error <1°), which solves the problem of misjudgment of the upper and lower levels of the overpass by traditional positioning systems (such as distinguishing the driving mileage of the main road / auxiliary road).

[0052] In medical settings and in vehicle monitoring, the positioning accuracy of mountain roads is enhanced by utilizing base station signals, and dual-antenna directional data is used to determine whether the vehicle deviates from the preset rescue route (such as triggering an early warning when entering an unpaved road).

[0053] Step S104. Obtain the high-frequency error and low-frequency error generated during the positioning process, correct the fused positioning data according to the high-frequency error and low-frequency error, and generate the vehicle positioning result according to the corrected fused positioning data.

[0054] Specifically, positioning errors are separated through frequency domain analysis: high-frequency errors (>1Hz, such as IMU zero bias, instantaneous wheel slippage) are corrected using real-time Kalman filtering; low-frequency errors (≤1Hz, such as satellite orbit errors, ionospheric delay) are compensated by extracting trend terms through sliding window Fourier transform and performing polynomial fitting.

[0055] Error frequency characteristics are divided into: high-frequency error sources: IMU angular rate noise (noise density is determined by Allan variance analysis) and odometer pulse counting error (isolated pulses are removed by median filtering); low-frequency error sources: satellite clock drift (predicted by differential data from the base station) and multipath effect accumulation (detected by dual-antenna phase differential consistency).

[0056] The differentiated correction strategy includes: high frequency correction: by designing a two-stage Kalman filter, the first stage processes high frequency noise from the IMU and odometry (update period 10ms), and the second stage fuses low frequency information from the satellite / dual antenna (update period 100ms). In financial scenarios, long-term drift can be corrected by fitting the trend term of the trajectory within a sliding window (5 seconds) 30 seconds before and after an accident (e.g., the cumulative error in the tunnel is reduced from 10 meters in the traditional solution to within 1 meter). In medical settings, by enabling a low-frequency error prediction model (trained based on historical 3-minute error data) during signal loss, positioning trends within tunnels can be predicted (such as inertial drift compensation caused by slope and curves).

[0057] In financial scenarios, low-frequency error correction ensures the accuracy of trajectory splicing at different time periods (e.g., position jumps of less than 0.3 meters before and after crossing a tunnel) in determining liability for accidents, thus avoiding disputes over liability due to accumulated errors.

[0058] In medical settings, during emergency rescue operations, high-frequency corrections suppress IMU errors caused by vehicle bumps in real time (such as reducing heading fluctuations from ±5° to ±1° during sharp turns), ensuring that the rescue dispatch system obtains accurate real-time location information.

[0059] In some embodiments, the BeiDou satellite signals include BeiDou-3 new frequency signals; such as Figure 2 As shown, the step of performing multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition includes steps S101a to S101b.

[0060] Step S101a. The BeiDou-3 new frequency signal is divided into frequency bands according to the corresponding BeiDou-3 new frequency point. The energy of the BeiDou-3 new frequency signal in each frequency band is accumulated. The accumulation time is dynamically adjusted according to the signal strength to obtain the accumulation result.

[0061] Step S101b. Weighted merge of the accumulated results for each frequency band to obtain the enhanced BeiDou satellite signal.

[0062] For signals from new BeiDou-3 frequency points (such as B1C, B2a, and B2b), the ability to acquire weak signals is improved through frequency band energy accumulation. The core steps are: frequency band processing → dynamic time accumulation → weighted merging, which solves the signal attenuation problem in complex environments.

[0063] Frequency band division and dynamic accumulation are achieved by dividing the frequency bands according to the new frequency points: B1C (1575.42MHz±4.092MHz), B2a (1176.45MHz±14.04MHz), and B2b (1207.14MHz±24MHz), with independent filtering for each band; dynamic adjustment of accumulation time includes: 50ms accumulation time when signal strength > -150dBm, 100ms when -150~-160dBm, and 200ms when < -160dBm (minimum 250ms accumulation is supported for medical scenarios); energy accumulation formula: Where It,Qt are the in-phase / quadrature components of the baseband signal, and T is the number of accumulation cycles.

[0064] The weighted merging strategy includes: weight coefficients set according to frequency band characteristics: B1C (high sensitivity, weight 0.4), B2a (anti-multipath, weight 0.3), and B2b (strong penetration, weight 0.3); in financial scenarios, priority is given to ensuring the continuity of the B1C frequency band for high-frequency positioning during accidents (10Hz update); in medical scenarios, the weight of the B2b frequency band is increased to 0.5 to enhance signal penetration capability in tunnels (cumulative signal acquisition rate increased by 30%).

[0065] In financial scenarios, in urban high-rise "canyon" areas, by combining and accumulating the B1C and B2a frequency bands, the positioning accuracy is improved from 5 meters in the traditional solution to 1.5 meters, ensuring that the coordinate error of the collision point in an accident is less than 1 meter, and reducing trajectory disputes in claims settlement.

[0066] In medical settings, within long tunnels (>2 km), the B2b band can be used to accumulate signals for 200ms for an extended period to maintain effective tracking of at least four satellites, preventing the vehicle-mounted monitoring system from being interrupted due to signal loss and thus gaining valuable time for emergency rescue.

[0067] In some embodiments, such as Figure 3As shown, before acquiring the inertial navigation data collected by the inertial navigation device mounted on the vehicle, the mileage data collected by the wheel odometer, and the orientation data collected by the dual antenna module, steps S105 and S106 are also included.

[0068] Step S105. Suppress electromagnetic interference on the BeiDou satellite signal by using a preset adaptive anti-narrowband interference circuit. By monitoring the signal spectrum in real time, identify the frequency range of narrowband interference and dynamically generate a notch filter corresponding to the interference frequency.

[0069] Step S106. The BeiDou satellite signal is filtered by the notch filter to remove narrowband interference components.

[0070] By identifying narrowband interference (such as stray signals from vehicle electronic devices and FM radio) through real-time spectrum monitoring, notch filters are dynamically generated to suppress interference and improve signal purity.

[0071] Interference detection and filter generation include: a spectrum monitoring module that scans the 0-2GHz band once per second with a resolution of 100kHz, and identifies narrowband interference bands through an energy threshold (>-100dBm); Notch filter design: For the interference center frequency f0, generate an IIR notch filter with a bandwidth of 2MHz, and a transfer function H(z) = (1 - 2rcosθz) / (1-2rcosθz). -1 +r 2 z -2 ) / (1-2cosθz -1 +z -2 ), where θ=2πf0 / fs, r=0.95 (suppression depth>30dB).

[0072] In financial scenarios, the focus is on monitoring 1560-1570MHz (GPS L1 band neighborhood interference), with an interference response time of <50ms, to ensure that mileage statistics during UBI billing are not interfered with by vehicle Bluetooth / Wi-Fi signals; in medical scenarios, monitoring of 400-500MHz (railway / industrial walkie-talkie band) is added, with a filter switching delay of <20ms, to avoid continuous interference of walkie-talkie signals on positioning during mountain rescue operations.

[0073] In financial settings, in areas with dense electronic devices such as parking lots, notch filtering improves the signal-to-noise ratio by 5dB, resolving the position jump problem caused by RFID reader interference in traditional positioning systems and ensuring the coordinate accuracy of vehicle entry and exit records (error < 2 meters). In medical settings, when ambulances pass through hospital MRI equipment areas, the 123-128MHz (MRI radio frequency interference band) is suppressed in real time to maintain stable positioning signals and avoid navigation errors in the rescue route caused by interference (such as missing emergency exits).

[0074] In some embodiments, before the satellite positioning data, inertial navigation data, mileage data, and orientation data are input into a preset compactly combined positioning model for fusion processing, the method further includes: if no base station signal is received, entering a base station-free positioning mode to process the BeiDou satellite signal using single-frequency precise single-point positioning technology to obtain satellite positioning data; and inputting the satellite positioning data, inertial navigation data, mileage data, and orientation data into the compactly combined positioning model.

[0075] When the base station signal is lost, single-frequency precise ephemeris (IGS fast orbit, 30-minute delay) and clock difference products (accuracy 0.1ns) are used to calculate decimeter-level positioning results through PPP technology, which are then used as inputs for the compact combination model.

[0076] The single-frequency PPP solution process includes: Error model: considering the first-order ionospheric term (corrected by the Klobuchar model), tropospheric delay (Saastamoinen model), and satellite clock bias (IGS final clock bias interpolation); State vector: including position (3D) and receiver clock bias (1D), using sequential least squares estimation (update rate 1Hz). In financial scenarios, almanac-assisted ephemeris prediction (predicting the orbit for the next 10 minutes with an error of <5 meters) ensures continuous location tracking in suburban areas. In medical scenarios, pre-stored 7-day IGS precise ephemeris data allows for offline calculation in remote mountainous areas without network coverage, maintaining positioning accuracy within 5 meters (compared to traditional solutions with an error >20 meters). The base station signal criterion triggers a switchover if no RTCM data is received for three consecutive cycles (3 seconds). In medical scenarios, an inertial navigation error threshold is added (forced switchover when the velocity error >5 m / s).

[0077] In financial scenarios, when the base station signal coverage is weak on remote sections of highways, single-frequency PPP is used to maintain positioning accuracy up to 3 meters, ensuring the continuity of the location when tracking stolen vehicles and avoiding tracking loss due to signal interruption.

[0078] In medical settings, in mountainous areas without base station coverage, the fusion of pre-stored ephemeris and inertial navigation reduces the positioning error for emergency rescue from 10 meters in traditional methods to 5 meters, providing more accurate landing coordinate references for helicopter rescue.

[0079] In some embodiments, the step of inputting satellite positioning data, inertial navigation data, odometer data, and orientation data into a preset compactly combined positioning model for fusion processing and outputting fused positioning data includes: establishing a state-space model of satellite positioning data, inertial navigation data, odometer data, and orientation data in the compactly combined positioning model; obtaining the velocity and heading corresponding to the inertial navigation data; predicting the vehicle position based on the inertial navigation data, using the velocity and heading as state variables, and using the position information corresponding to the satellite positioning data and the displacement information corresponding to the odometer data as observation variables; iteratively updating the state variables using a Kalman filter algorithm; and introducing dual-antenna orientation data to correct and constrain the heading angle, and outputting the fused positioning data.

[0080] By constructing a 15-dimensional state-space model (position, velocity, attitude, IMU error, clock error, etc.), using inertial navigation to predict the state, satellite positioning and odometer as observations, and introducing dual-antenna orientation data to correct the heading angle, the heading drift problem under the viaduct is solved.

[0081] The state equations corresponding to the state-space model construction include: Where F is the system matrix (including Earth's rotation and IMU error model), and G is the noise matrix; observation equations: satellite pseudorange / carrier phase residual, odometer displacement difference, dual-antenna heading angle difference, with barometric altimeter observation added for medical scenarios (to improve elevation accuracy); dual-antenna constraints: heading angle observation equation zθ=θdual-θimu+vθ, measurement noise covariance set to 0.01°. 2 (Financial scenario) / 0.005° 2 (Medical scenario).

[0082] In financial scenarios, before an accident collision, a "high-gain filtering" mode is triggered (the process noise covariance is increased by 2 times) to quickly respond to sudden position changes (such as increasing the position update rate to 20Hz during emergency braking). In medical settings, a "heading lock" mechanism is activated inside the tunnel. When the dual-antenna signals are stable, the heading angle update weight is forcibly increased to 0.8 to suppress IMU drift (heading error is reduced from ±5° to ±1°).

[0083] In financial scenarios, when driving on different levels of an elevated highway, the dual-antenna heading constraint accurately distinguishes between the main road and the auxiliary road (for example, the distance between the main and auxiliary roads on the West Second Ring Road in Beijing is 30 meters, and the misjudgment rate of the traditional solution is 20%, while this solution reduces it to 1%), ensuring the accuracy of UBI billing mileage.

[0084] In medical settings, when driving on mountain curves, dual antennas are used to correct the course in real time. Combined with high-frequency sampling (200Hz) by IMU, the vehicle's turning angle error is reduced from ±10° to ±3°, helping the rescue system determine whether the vehicle has deviated from the safe route (such as issuing a warning when entering a cliff section).

[0085] In some embodiments, acquiring the high-frequency and low-frequency errors generated during the positioning process includes: classifying the errors during the positioning process according to their time characteristics; classifying receiver clock error and signal acquisition noise as the high-frequency errors; estimating the high-frequency errors periodically using real-time satellite signal observations; classifying ionospheric delay and tropospheric delay as the low-frequency errors; and predicting the trend of the low-frequency errors based on historical observation data from the reference station network and real-time meteorological data.

[0086] By separating errors according to their time characteristics: high-frequency errors (<1 second period, such as receiver clock error, multipath noise) are estimated periodically; low-frequency errors (>10 second period, such as slow changes in the ionosphere) are predicted in terms of trend, thereby improving the targeting of error correction.

[0087] Error classification and estimation methods include: High-frequency errors (100Hz update rate in financial scenarios, 200Hz in medical scenarios): Receiver clock bias: estimated using the satellite common-view method, error <5ns (corresponding to 1.5m pseudorange error); Signal acquisition noise: using pseudorange difference between adjacent epochs, setting a 3σ threshold to remove outliers (0.5m threshold in financial scenarios, 0.3m threshold in medical scenarios); Low-frequency errors: Ionospheric delay: based on a 15-minute moving average model of the base station network, generating a regional ionospheric grid (grid spacing 50km, densified to 20km in medical scenarios); Tropospheric delay: combined with real-time air pressure / temperature data (collected by vehicle-mounted sensors in medical scenarios, and called from meteorological APIs in financial scenarios), corrected using the Saastamoinen model.

[0088] In financial scenarios, the focus is on monitoring ionospheric abrupt changes during the day-to-day period (such as the rate of change of delay at sunrise > 10 TECU / min) to trigger low-frequency error re-estimation and ensure the consistency of the trajectory throughout the day. In medical settings, during severe weather such as heavy rain, the tropospheric correction weight is increased to 0.6 (default 0.4), and by utilizing real-time data from the vehicle-mounted barometer (accuracy ±0.5 hPa), the elevation error is reduced from 5 meters to 2 meters.

[0089] In financial scenarios, during cross-day and night accident handling, low-frequency error trend prediction can be used to correct trajectory drift caused by diurnal variations in the ionosphere (e.g., the error drops from 8 meters to 1.5 meters at 2 a.m.), providing a continuous and accurate coordinate sequence across time periods for accident liability determination.

[0090] In medical settings, during cloudy weather in mountainous areas, the system estimates tropospheric delay changes in real time (updated every 20 seconds) to avoid excessive positioning elevation errors caused by clouds and fog (the traditional solution has an elevation error of >10 meters, while this solution has an error of <3 meters), ensuring accurate altitude determination for rescue helicopters.

[0091] In some embodiments, the step of correcting the fused positioning data based on the high-frequency and low-frequency errors to generate a vehicle positioning result based on the corrected fused positioning data includes: applying a real-time Kalman filter algorithm to filter and correct the high-frequency errors, compensating them into the position, velocity, and heading parameters corresponding to the fused positioning data; and for the low-frequency errors, by receiving low-frequency error correction information for regular grid points broadcast by the reference station network, matching the correction values ​​of neighboring grid points according to the vehicle's real-time position, and weighting and correcting the ionospheric and tropospheric delay errors in the fused positioning data to generate the vehicle positioning result containing position coordinates and heading angle.

[0092] High-frequency errors are compensated in real time through Kalman filtering, while low-frequency errors are corrected using weighted correction information from the reference station network, thus achieving precise error suppression in dynamic scenarios.

[0093] High-frequency error correction (100Hz real-time processing) includes: the state vector contains high-frequency error terms (clock drift rate, IMU zero bias); dynamic adjustment of the Kalman filter gain matrix: in financial scenarios, the position error covariance is increased during rapid acceleration (gain +30%); in medical scenarios, the inertial navigation weight is pre-increased before signal loss (gain +50%); the correction formula includes: , where K is the Kalman gain, which is compensated in real time for position, velocity, and heading parameters.

[0094] Low-frequency error correction (1Hz grid matching) includes: low-frequency error (ionospheric / tropospheric) of grid points published by the base station network, with a grid resolution of 50km×50km for financial scenarios and 20km×20km for medical scenarios; matching the real-time location of the vehicle with four neighboring grid points, and calculating the correction value using bilinear interpolation: , where wi is the distance weighting coefficient (altitude weight is added in medical scenarios).

[0095] In financial scenarios, when reconstructing insurance claim trajectories, low-frequency error grid correction can reduce long-term drift across urban areas (such as cumulative error over 1 hour) from 20 meters to less than 3 meters, ensuring global consistency of positioning results across different road segments and avoiding disputes over liability determination due to error accumulation.

[0096] In medical settings and emergency rescue dispatch, high-frequency correction suppresses IMU noise caused by vehicle bumps in real time (such as reducing speed error from 2m / s to 0.5m / s during emergency braking). Combined with low-frequency grid correction, the real-time positioning accuracy is improved to a plane error of <1.5 meters and an elevation error of <2.5 meters, meeting the coordinate requirements for precise airdrop of rescue supplies by helicopter.

[0097] In some embodiments, a generative adversarial network (GAN) is constructed to address the feature loss problem when BeiDou signals are blocked. The generator restores the phase / amplitude features of weak signals, and the discriminator distinguishes between real signals and generated signals, thereby improving the signal acquisition success rate in low signal-to-noise ratio (<-165dBm) scenarios.

[0098] The network architecture design includes: Generator (G): 3 layers of transposed convolution (input layer 100-dimensional random noise → output layer generates I / Q baseband signal, size 1024×2), activation function is ReLU (the last layer tanh normalizes to [-1,1]); Discriminator (D): 3 layers of convolution (input real / generated signal → output 0-1 probability value), loss function is WGAN-GP (Wasserstein distance + gradient penalty), training data comes from real-world -160~-170dBm weak signal samples (100,000 sets each collected in financial / medical scenarios).

[0099] The signal enhancement process includes: Preprocessing: Downconverting the original signal to baseband and extracting a 1ms data segment as input (B1C frequency for financial scenarios, B2b frequency for medical scenarios); Generation enhancement: When the measured signal-to-noise ratio is <-160dBm, triggering GAN to generate a compensation signal, which is superimposed on the original signal at an energy ratio of 0.3:0.7 (focusing on the real signal for financial scenarios, increasing the weight of the generated signal to 0.5 for medical scenarios); Acquisition verification: Verifying the enhanced signal through parallel code phase search, and updating the GAN parameters after successful acquisition (online incremental learning).

[0100] In the financial scenario, at the entrance of an underground parking lot (signal strength -163dBm), GAN enhancement reduces the signal acquisition time from 800ms in the traditional solution to 300ms, avoiding billing errors caused by positioning delays when vehicles enter and exit (such as reducing the probability of missing an entry or exit record from 5% to 0.5%).

[0101] In medical scenarios, in dense forest areas (signal strength -168dBm), the generator prioritizes the restoration of multipath fading characteristics of the B2b frequency point. With the assistance of inertial navigation prediction, the signal reacquisition success rate is increased from 20% to 70%, ensuring that ambulances can report their real-time location in areas without obvious road signs (delay <1 second).

[0102] In some embodiments, a deep reinforcement learning (DRL) model is constructed, with the signal carrier-to-noise ratio (CN0) and Doppler frequency shift rate as state inputs, and the optimal accumulation time (T) and frequency band weight (W) as outputs, replacing the traditional fixed threshold strategy, to achieve adaptive accumulation in complex dynamic scenarios.

[0103] The DRL architecture design includes: State space (S): [current CN0, CN0 change rate in the previous 3 seconds, satellite elevation angle, vehicle acceleration] (4-dimensional); Action space (A): Accumulation time T∈{50,100,200,300ms}, frequency band weight W∈{[0.5,0.3,0.2],[0.3,0.4,0.3], [0.2,0.2,0.6]} (corresponding to B1C / B2a / B2b); Reward function (R): R=+100 for successful capture, R=-50 for uncaptured and CN0 decrease. The PPO algorithm is used for training, and the experience replay pool has a capacity of 100,000 records. Data is collected in urban canyons and mountain tunnels for financial and medical scenarios, respectively.

[0104] The online decision-making process includes: collecting state S every 200ms and inputting it into the DRL model to generate action A; after executing the action, calculating the new CN0 and capturing the state as feedback for the next state, and updating the policy network (the update frequency is increased to 100ms in the medical scenario); setting an "emergency braking protection" action in the financial scenario: when acceleration > 0.5g is detected, T = 100ms + W = [0.5, 0.3, 0.2] is forcibly selected to ensure signal stability at the moment of the accident; setting a "tunnel mode" priority in the medical scenario: when GPS signal loss is detected, the maximum weight of the B2b band is selected first (W = [0, 0, 1]).

[0105] In financial scenarios, such as frequent lane changes on urban expressways (CN0 fluctuation ±3dB), the DRL model reduces the cumulative parameter adjustment delay from 500ms in the traditional scheme to 150ms, while maintaining a positioning update rate of 10Hz, ensuring the accuracy of coordinate points for sudden acceleration / sudden braking events in UBI insurance (error < 1.2 meters).

[0106] In medical scenarios, when driving in long tunnels (>3 km), the DRL model automatically adjusts the accumulation time to 300ms (the traditional solution has a maximum of 200ms) and dynamically allocates the weight of the B2b frequency band to 0.8, increasing the number of satellites tracked from 2 in the traditional solution to 4, thus avoiding misjudgment of the vehicle's position by the rescue dispatch system (such as falsely reporting that the vehicle has left the tunnel).

[0107] In some embodiments, for scenarios without a reference station, a global high-precision PPP model (pre-trained on IGS global station data) is transferred to a regional scenario using transfer learning technology. Fine-tuning is then performed using a small amount of local data (100 sets / scenario) to improve the single-frequency positioning accuracy in complex terrain.

[0108] The transfer learning framework includes: a pre-trained model: a 12-layer Transformer network built on TensorFlow, with inputs of satellite ephemeris parameters, ionospheric model parameters, and receiver clock priors, and output of position coordinates (3D); domain adaptation: 100 sets of labeled data (RTK ground truth) were collected in the hilly areas of the Yangtze River Delta for the financial scenario and in the mountainous areas of Sichuan and Yunnan for the medical scenario, the first 8 layers of the pre-trained model were frozen, and the last 4 layers were fine-tuned (learning rate 1e-4); and an error compensation module: a terrain height correction branch was added (inputting SRTM 90m resolution elevation data), with the financial scenario focusing on plane error (weight 0.7) and the medical scenario taking into account elevation (weight 0.5).

[0109] The inference optimization strategy includes: in the financial scenario, in the bridge / elevation scenario, enabling "structural feature transfer" (inputting the bridge coordinate prior library) reduces the planar positioning error from 4 meters to 2 meters, solving the positional offset caused by the multipath effect in traditional PPP (such as misjudging lanes); in the medical scenario, in areas with an altitude > 2000 meters, introducing barometric altimeter data (accuracy ±1m) as an auxiliary input for the transfer model reduces the elevation positioning error from 8 meters to 3 meters, meeting the elevation accuracy requirements of helicopter take-off and landing points in mountainous areas.

[0110] In financial scenarios, in the dense urban cluster of the Pearl River Delta (where high-rise buildings block more than 40% of the view), the PPP model after transfer learning improves the positioning accuracy without a base station from 8 meters in the traditional solution to 3 meters, ensuring the accuracy of trajectory reconstruction of stolen vehicles in offline state and providing effective tracking coordinates.

[0111] In medical settings, in the Hengduan Mountains (elevation difference > 1000 meters / 10 kilometers), by migrating terrain features and fusing air pressure data, the elevation error of single-frequency PPP can be reduced from 15 meters to 5 meters, avoiding altitude control errors of rescue helicopters (such as the risk of flying too low) caused by elevation errors.

[0112] In some embodiments, a graph neural network model is constructed to represent satellite positioning, inertial navigation, odometer, and dual-antenna data as graph nodes, with edge weights reflecting data correlation (such as time synchronization and sensor accuracy). The optimal fusion strategy in a dynamic environment is achieved through graph convolution operations.

[0113] The graph structure definition includes: Node features (N): Satellite nodes (CN0, elevation angle, azimuth angle), IMU nodes (acceleration, angular velocity, zero bias error), odometry nodes (wheel speed, cumulative mileage), dual-antenna nodes (baseline length, heading angle variance); Edge features (E): timestamp synchronization difference (<5ms indicates strong connection), sensor spatial location correlation (distance <2m indicates strong connection), and the adjacency matrix A calculates the similarity between nodes through dynamic time warping (DTW); Graph Convolutional Layer (GCN): adopts GAT (Graph Attention Mechanism), focusing on satellite node weights (4 heads) in financial scenarios and IMU node weights (6 heads) in medical scenarios.

[0114] The fusion decision-making mechanism includes: State update: Graph convolution is performed every 10ms to output fused position / velocity / heading estimates; Anomaly detection: When satellite node CN0 < -155dBm and IMU node angular velocity variance > 0.1° 2 In the financial scenario, an "inertia-driven" mode is triggered (increasing the IMU weight to 0.6). An "accident feature node" is added: when a triaxial acceleration > 0.3g is detected, all sensor timestamps are forcibly synchronized (delay < 1ms) to ensure data fusion accuracy at the moment of collision. In the medical scenario, a "rescue priority node" is added: when an emergency signal is received, the weight of the dual-antenna node is increased to 0.7 to suppress interference from abnormal data from other sensors. In the financial scenario, in situations where multiple sensors are out of sync (e.g., odometer signal delay of 10ms), the GNN model dynamically adjusts the weights through side features, reducing the fusion positioning error from 2.5 meters in the traditional EKF to 1.2 meters, ensuring the accuracy of key coordinate points (e.g., braking start point) in accident liability determination. In the medical scenario, when an ambulance traverses multiple tunnels (intermittent signal scenario), the GNN model automatically identifies satellite node failure states, increasing the sum of the IMU and odometer weights from 0.5 to 0.8 to maintain positioning continuity (position drift < 5 meters / minute during signal loss), providing real-time location reference for remote medical guidance.

[0115] In some embodiments, by using a Long Short-Term Memory (LSTM) network to model the temporal correlation of high-frequency errors (receiver clock error, multipath noise), the error at the next moment is predicted by using a historical 10-second error sequence, replacing the traditional cycle-by-cycle estimation, thus improving the timeliness of correction in dynamic scenarios.

[0116] The LSTM model design includes: Input layer: clock error residuals, pseudorange multipath error, and carrier phase noise (3D sequence) from the past 10 epochs (100ms); Hidden layer: 2 LSTM layers (128 units each) + 1 fully connected layer, activation function tanh, loss function MAE, 500,000 error samples were collected in urban congestion and mountainous sharp bend environments for financial and medical scenarios respectively; Prediction step size: 50ms error prediction for the financial scenario (20Hz update rate), and 20ms error prediction for the medical scenario (50Hz update rate).

[0117] The error correction process includes: online prediction: when a vehicle acceleration > 0.2g is detected (dynamic scenario), the LSTM prediction mode is enabled, and the traditional cycle-by-cycle estimation is used as a backup; correction fusion is achieved by weighting the prediction error and the real-time estimation error according to weights (0.7:0.3 for dynamic scenarios, 0.3:0.7 for static scenarios), and a "sudden braking trigger threshold" is set for financial scenarios (the prediction weight is increased to 0.9 when the acceleration > 0.5g); "physiological signal correlation" is added for medical scenarios: when the vehicle monitor detects abnormal vital signs of the patient, the LSTM model is forced to use the highest prediction frequency (100Hz) to ensure that the positioning update is synchronized with the rescue response.

[0118] In financial scenarios and in urban congestion with frequent start-stop operations, the LSTM model reduces the receiver clock error prediction error from 8ns in the traditional solution to 3ns (corresponding to a pseudorange error of 2.4 meters to 0.9 meters), solving the position jump problem caused by sudden changes in clock error when following other vehicles (such as reducing the number of misjudged lane changes by 30%).

[0119] In medical settings, on mountainous roads with sharp bends (curvature radius < 50 meters), LSTM is used to predict multipath noise, reducing the heading angle fluctuation from ±3° to ±1.5°. Combined with inertial navigation correction, this ensures the real-time heading accuracy of ambulances when driving on narrow mountain roads (error < 2°), preventing the navigation system from misjudging the driving direction (such as mistakenly pointing to the wrong direction).

[0120] In some embodiments, by constructing a federated learning framework, multiple base stations and vehicle-mounted terminals within the region are united (financial / medical terminals are independently networked), and low-frequency error (ionospheric / tropospheric) grid models are dynamically updated while protecting data privacy, thus solving the problem of update delay in traditional centralized models.

[0121] The federated learning architecture includes: Server-side: Maintaining a global low-frequency error grid model (initialized as an IGS regional model), with separate servers for financial and medical scenarios to prevent data overlap; Client-side: Vehicle terminals periodically (every 10 minutes for financial scenarios, every 5 minutes for medical scenarios) upload local observation residuals (anonymized, containing only latitude and longitude + residual mean), without transmitting raw signal data; Model aggregation: Using the FedAvg algorithm, the weight of financial scenarios focuses on urban area clients (accounting for 60%), and the weight of medical scenarios focuses on suburban / mountainous clients (accounting for 70%), with an aggregation cycle of 10 rounds.

[0122] In financial scenarios, in densely populated commercial areas (such as Lujiazui in Shanghai), high-frequency client data (1000+ vehicles / area) is used to improve the ionospheric grid resolution from 50km to 10km. After correction, the planar error is reduced from 4 meters to 1.8 meters, ensuring accurate differentiation of vehicle positions in parking lots (e.g., 2.5 meters between adjacent parking spaces, with a positioning error of <1 meter). In the medical setting, a micro-federal network (50+ ambulances + 3 temporary base stations) is established in mountainous areas. When heavy rain is detected, emergency aggregation is triggered (the cycle is shortened to 1 minute), and a tropospheric correction grid for the rain-affected area is dynamically generated, reducing the elevation error from 10 meters to 4 meters, thus ensuring the accuracy of airdropping relief supplies in mountainous areas.

[0123] In financial scenarios, federated learning models update the ionospheric grid of high-density urban areas every hour, solving the problem of traditional RTCM broadcast delay (>30 seconds), making the real-time mileage statistics error of UBI insurance <0.1%, and improving the credibility of the billing system.

[0124] In medical settings, within quarantine areas, vehicle-mounted terminals upload anonymized positioning residuals through federated learning, assisting disease control centers in dynamically updating low-frequency error models within the area. This ensures that the positioning accuracy of ambulances in the lockdown area is not affected by changes in the electromagnetic environment (such as interference from temporarily added 5G base stations), guaranteeing the real-time accuracy of transport route planning.

[0125] Please see Figure 4 As shown, Figure 4 This is a schematic diagram of the structure of a vehicle-mounted positioning device 200 based on BeiDou satellite signals provided in this application embodiment. The vehicle-mounted positioning device 200 based on BeiDou satellite signals is used to execute the steps of the vehicle-mounted positioning method based on BeiDou satellite signals shown in the above embodiments. The vehicle-mounted positioning device 200 based on BeiDou satellite signals can be a single server or a server cluster, or it can be a terminal, such as a handheld terminal, a laptop computer, a wearable device, or a robot.

[0126] like Figure 4As shown, the vehicle-mounted positioning device 200 based on BeiDou satellite signals includes: The signal acquisition unit 201 is used to acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition. The data acquisition unit 202 is used to acquire inertial navigation data collected by the inertial navigation device mounted on the vehicle, mileage data collected by the wheel odometer, and orientation data collected by the dual antenna module; The fusion output unit 203 is used to enter the base station positioning mode when a base station signal is received, process the Beidou satellite signal according to the single-frequency real-time dynamic differential technology, obtain satellite positioning data, input the satellite positioning data, inertial navigation data, mileage data and orientation data into a preset compact combination positioning model for fusion processing, and output fused positioning data. The result generation unit 204 is used to acquire the high-frequency error and low-frequency error generated during the positioning process, correct the fused positioning data according to the high-frequency error and low-frequency error, and generate the vehicle positioning result according to the corrected fused positioning data.

[0127] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the vehicle-mounted positioning device and its modules based on BeiDou satellite signals described above can be referred to the corresponding processes in the vehicle-mounted positioning method embodiments based on BeiDou satellite signals described above, and will not be repeated here.

[0128] The aforementioned vehicle positioning method based on BeiDou satellite signals can be implemented as a computer program, which can be used in various applications such as... Figure 4 It runs on the device shown.

[0129] Please see Figure 5 , Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. The computer device includes a processor, a memory, and a network interface connected via a device bus, wherein the memory may include a storage medium and internal memory.

[0130] The storage medium can store operating devices and computer programs. The computer program includes program instructions that, when executed, cause the processor to perform any vehicle positioning method based on BeiDou satellite signals.

[0131] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0132] The internal memory provides an environment for the execution of computer programs stored in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any vehicle positioning method based on BeiDou satellite signals.

[0133] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the terminal to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0134] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0135] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: Acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition; Acquire inertial navigation data collected by the vehicle's inertial navigation equipment, mileage data collected by the wheel odometer, and orientation data collected by the dual-antenna module; If a base station signal is received, the system enters the base station positioning mode. The BeiDou satellite signal is processed according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. The satellite positioning data, inertial navigation data, mileage data and orientation data are input into a preset compact combination positioning model for fusion processing, and the fused positioning data is output. The high-frequency and low-frequency errors generated during the positioning process are acquired, and the fused positioning data is corrected based on the high-frequency and low-frequency errors. The vehicle positioning result is then generated based on the corrected fused positioning data.

[0136] In some embodiments, the BeiDou satellite signal includes BeiDou-3 new frequency point signals; the step of performing multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signal to enhance weak signal acquisition includes: dividing the BeiDou-3 new frequency point signals into frequency bands according to the corresponding BeiDou-3 new frequency points, accumulating the energy of the BeiDou-3 new frequency point signals in each frequency band, and dynamically adjusting the accumulation time according to the signal strength to obtain the accumulation result; and weighting and merging the accumulation results of each frequency band to obtain the enhanced BeiDou satellite signal.

[0137] In some embodiments, before acquiring the inertial navigation data collected by the inertial navigation device mounted on the vehicle, the mileage data collected by the wheel odometer, and the orientation data collected by the dual-antenna module, the method further includes: suppressing electromagnetic interference of the BeiDou satellite signal through a preset adaptive anti-narrowband interference circuit; identifying the frequency range of narrowband interference by real-time monitoring of the signal spectrum; dynamically generating a notch filter corresponding to the interference frequency; and filtering the BeiDou satellite signal through the notch filter to remove narrowband interference components.

[0138] In some embodiments, before the satellite positioning data, inertial navigation data, mileage data, and orientation data are input into a preset compactly combined positioning model for fusion processing, the method further includes: if no base station signal is received, entering a base station-free positioning mode to process the BeiDou satellite signal using single-frequency precise single-point positioning technology to obtain satellite positioning data; and inputting the satellite positioning data, inertial navigation data, mileage data, and orientation data into the compactly combined positioning model.

[0139] In some embodiments, the step of inputting satellite positioning data, inertial navigation data, odometer data, and orientation data into a preset compactly combined positioning model for fusion processing and outputting fused positioning data includes: establishing a state-space model of satellite positioning data, inertial navigation data, odometer data, and orientation data in the compactly combined positioning model; obtaining the velocity and heading corresponding to the inertial navigation data; predicting the vehicle position based on the inertial navigation data, using the velocity and heading as state variables, and using the position information corresponding to the satellite positioning data and the displacement information corresponding to the odometer data as observation variables; iteratively updating the state variables using a Kalman filter algorithm; and introducing dual-antenna orientation data to correct and constrain the heading angle, and outputting the fused positioning data.

[0140] In some embodiments, acquiring the high-frequency and low-frequency errors generated during the positioning process includes: classifying the errors during the positioning process according to their time characteristics; classifying receiver clock error and signal acquisition noise as the high-frequency errors; estimating the high-frequency errors periodically using real-time satellite signal observations; classifying ionospheric delay and tropospheric delay as the low-frequency errors; and predicting the trend of the low-frequency errors based on historical observation data from the reference station network and real-time meteorological data.

[0141] In some embodiments, the step of correcting the fused positioning data based on the high-frequency and low-frequency errors to generate a vehicle positioning result based on the corrected fused positioning data includes: applying a real-time Kalman filter algorithm to filter and correct the high-frequency errors, compensating them into the position, velocity, and heading parameters corresponding to the fused positioning data; and for the low-frequency errors, by receiving low-frequency error correction information for regular grid points broadcast by the reference station network, matching the correction values ​​of neighboring grid points according to the vehicle's real-time position, and weighting and correcting the ionospheric and tropospheric delay errors in the fused positioning data to generate the vehicle positioning result containing position coordinates and heading angle.

[0142] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the steps of the vehicle positioning method based on BeiDou satellite signals as described in the first aspect above.

[0143] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0144] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vehicle-mounted positioning method based on BeiDou satellite signals, characterized in that, include: Acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition; Acquire inertial navigation data collected by the vehicle's inertial navigation equipment, mileage data collected by the wheel odometer, and orientation data collected by the dual-antenna module; If a base station signal is received, the system enters the base station positioning mode. The BeiDou satellite signal is processed according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. The satellite positioning data, inertial navigation data, mileage data and orientation data are input into a preset compact combination positioning model for fusion processing, and the fused positioning data is output. The high-frequency and low-frequency errors generated during the positioning process are acquired, and the fused positioning data is corrected based on the high-frequency and low-frequency errors. The vehicle positioning result is then generated based on the corrected fused positioning data.

2. The method according to claim 1, characterized in that, The BeiDou satellite signals include BeiDou-3 new frequency signals; the multi-frequency noncoherent accumulation processing of the acquired BeiDou satellite signals to enhance weak signal acquisition includes: The BeiDou-3 new frequency signal is divided into frequency bands according to the corresponding BeiDou-3 new frequency point. The energy of the BeiDou-3 new frequency signal in each frequency band is accumulated. The accumulation time is dynamically adjusted according to the signal strength to obtain the accumulation result. The accumulated results for each frequency band are weighted and merged to obtain the enhanced BeiDou satellite signal.

3. The method according to claim 1, characterized in that, Before acquiring the inertial navigation data collected by the vehicle-mounted inertial navigation device, the mileage data collected by the wheel odometer, and the orientation data collected by the dual-antenna module, the method further includes: Electromagnetic interference is suppressed on the BeiDou satellite signal by a preset adaptive anti-narrowband interference circuit. By monitoring the signal spectrum in real time, the frequency range of narrowband interference is identified, and a notch filter corresponding to the interference frequency is dynamically generated. The BeiDou satellite signal is filtered by the notch filter to remove narrowband interference components.

4. The method according to claim 1, characterized in that, Before the process of inputting satellite positioning data, inertial navigation data, odometer data, and orientation data into a preset compactly combined positioning model for fusion processing, the method further includes: If no base station signal is received, the system enters a base station-less positioning mode to process the BeiDou satellite signal using single-frequency precise single-point positioning technology to obtain satellite positioning data; the satellite positioning data, inertial navigation data, mileage data, and orientation data are then input into the tightly coupled positioning model.

5. The method according to claim 1, characterized in that, The process of inputting satellite positioning data, inertial navigation data, mileage data, and orientation data into a preset compact combination positioning model for fusion processing, and outputting fused positioning data, includes: In the compact combination positioning model, a state-space model is established for satellite positioning data, inertial navigation data, mileage data, and orientation data; Obtain the speed and heading corresponding to the inertial navigation data; The vehicle position, speed, and heading are predicted based on inertial navigation data as state variables, and the position information corresponding to satellite positioning data and the displacement information corresponding to mileage data are used as observation variables. The state variables are iteratively updated using a Kalman filter algorithm, and dual-antenna orientation data is introduced to correct and constrain the heading angle, and the fused positioning data is output.

6. The method according to claim 1, characterized in that, The acquisition of high-frequency and low-frequency errors generated during the positioning process includes: The errors in the positioning process are classified according to their time characteristics. Receiver clock error and signal acquisition noise are classified as high-frequency errors. The high-frequency errors are estimated cycle by cycle using real-time satellite signal observations. Ionospheric delay and tropospheric delay are classified as low-frequency errors, and the trend of low-frequency errors is predicted based on historical observation data from the reference station network and real-time meteorological data.

7. The method according to claim 1, characterized in that, The step of correcting the fused positioning data based on the high-frequency error and the low-frequency error, and generating a vehicle positioning result based on the corrected fused positioning data, includes: High-frequency errors are filtered and corrected using a real-time Kalman filter algorithm, and the compensation is added to the position, velocity, and heading parameters corresponding to the fused positioning data. For low-frequency errors, the system receives low-frequency error correction information from regular grid points broadcast by the reference station network, matches the correction values ​​of neighboring grid points based on the vehicle's real-time position, and performs weighted correction on the ionospheric and tropospheric delay errors in the fused positioning data to generate the vehicle positioning result containing position coordinates and heading angle.

8. A vehicle-mounted positioning device based on BeiDou satellite signals, characterized in that, include: The signal acquisition unit is used to acquire BeiDou satellite signals and perform multi-frequency noncoherent accumulation processing on the acquired BeiDou satellite signals to enhance weak signal acquisition. The data acquisition unit is used to acquire inertial navigation data collected by the inertial navigation device mounted on the vehicle, mileage data collected by the wheel odometer, and orientation data collected by the dual-antenna module. The fusion output unit is used to enter the base station positioning mode when a base station signal is received. It processes the BeiDou satellite signal according to the single-frequency real-time dynamic differential technology to obtain satellite positioning data. It inputs the satellite positioning data, inertial navigation data, mileage data and orientation data into a preset compact combination positioning model for fusion processing and outputs fused positioning data. The result generation unit is used to acquire high-frequency and low-frequency errors generated during the positioning process, correct the fused positioning data based on the high-frequency and low-frequency errors, and generate vehicle positioning results based on the corrected fused positioning data.

9. A computer device, characterized in that, The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the method as described in any one of claims 1 to 7.

Images