Low-cost intelligent terminal satellite and inertial combination precision single-point positioning method and low-cost intelligent terminal satellite and inertial combination precision single-point positioning device

Abstract

The invention discloses a low-cost intelligent terminal satellite and inertial integrated precise point positioning method and device, and belongs to the technical field of navigation positioning. The method comprises the following steps: aiming at the problem that the pseudo-range inter-frequency deviation of the low-cost terminal is large and gradually varied, carrying out deviation estimation and correction by adopting a semi-parameter estimation and compensation least square method; for observation value abnormity and deletion, RAIM and LSTM are fused to construct an early warning-recovery-verification closed loop mechanism; aiming at a dual-frequency and single-frequency observation value coexistence scene, a mixed ionosphere delay processing strategy is adopted, and differential correction is realized through ionosphere elimination combination and GIM constraint; aiming at the problem of combination switching lag, constructing a fuzzy-neural network adaptive decision model, and dynamically selecting a GNSS / INS combination mode based on multi-dimensional features; and position parameter estimation and post-processing are realized through an integrated algorithm framework. According to the invention, high-precision and high-robustness precise point positioning is realized under the condition of low-cost hardware.

Description

Technical Field

[0001] This invention belongs to the field of navigation and positioning technology, specifically relating to a low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method and device. Background Technology

[0002] With the rapid development of technologies such as the Internet of Things (IoT), wearable technology, and autonomous driving, the demand for high-precision location services from low-cost smart terminals, such as smartphones, IoT modules, and consumer drones, is becoming increasingly urgent. Traditional location, velocity, and timing technologies typically provide meter-level positioning accuracy, which is insufficient for high-precision application scenarios ranging from centimeters to decimeters. Precise point positioning (PPS) technology, which can achieve high-precision positioning globally without relying on ground reference stations, has become an important way to improve the positioning capabilities of low-cost terminals.

[0003] However, applying precise point positioning technology to low-cost smart terminals faces a series of severe challenges due to the inherent limitations of the terminal hardware. Limited by cost, power consumption, and size, such terminals are typically equipped with low-gain GNSS antennas and baseband chips with weak processing capabilities, resulting in a significant deterioration in the quality of raw satellite observations. Specifically: First, pseudo-range-frequency deviations are abnormally significant, reaching thousands to tens of thousands of nanoseconds, and exhibiting slow changes between epochs, far exceeding the processing capabilities of traditional models, easily leading to divergence in positioning solutions; second, observations are noisy, prone to cycle slips, clock slips, or complete data loss in complex environments, while existing recovery methods mostly rely on a single data source, lacking accuracy and early warning capabilities in cases of severe data loss; third, terminals often encounter mixed scenarios with both dual-frequency and single-frequency observations due to insufficient signal processing capabilities, making it impossible for traditional single ionospheric processing strategies to balance observation utilization and correction accuracy.

[0004] In addition, existing satellite navigation / inertial navigation combination strategies are mostly based on a single threshold of the number of available satellites for mode switching, failing to comprehensively consider key factors such as observation quality, inertial navigation drift, and environmental dynamics. In complex scenarios such as urban canyons and high-speed movement, switching lag is likely to occur, leading to a decrease in positioning continuity.

[0005] In summary, existing technologies are insufficient to systematically solve the core problems faced by low-cost smart terminals in achieving precise single-point positioning, such as "large deviation, poor data, rigid strategies, and sluggish switching." There is an urgent need for an integrated positioning solution that can adapt to the characteristics of low-cost hardware and maintain high accuracy and robustness in complex environments. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a low-cost intelligent terminal satellite-inertial navigation system combined precise single-point positioning method and device. Through collaborative innovation of "pseudorange deviation correction - observation anomaly handling - ionospheric delay adaptation - combination mode optimization - full-process integration", a solution with both technical and engineering practicality is formed.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A low-cost intelligent terminal satellite-inertial navigation system combined precise single-point positioning method, the method comprising:

[0009] Step 1: Acquire GNSS observation data and IMU data from a low-cost smart terminal;

[0010] Step 2: Based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation;

[0011] Step 3: Perform quality analysis on the corrected GNSS observation data. When abnormal or missing observation data is detected, use the mechanism of autonomous integrity monitoring RAIM and long short-term memory network LSTM for early warning and recovery.

[0012] Step 4: Based on the type of the recovered observations, construct an observation model for location calculation using a hybrid ionospheric delay processing strategy;

[0013] Step 5: Dynamically select the combination mode of GNSS and INS through an adaptive decision model that integrates fuzzy logic and neural networks;

[0014] Step 6: Based on the selected combination mode and the corresponding observation model, dynamically construct the system state equation using the IMU data, estimate the position parameters using extended Kalman filtering, and output the final positioning result after post-processing.

[0015] Furthermore, step 2 includes:

[0016] Calculate the initial value of ionospheric delay based on the broadcast ionospheric model or the Global Ionospheric Map (GIM) product;

[0017] Construct a pseudorange observation equation that defines the pseudorange-frequency deviation as a non-modeling error that needs to be estimated separately;

[0018] The expression for the inter-frequency deviation is derived from the pseudorange observation equation to obtain its initial value;

[0019] Construct constraint equations to limit the maximum change in frequency deviation between adjacent epochs;

[0020] A compensated least squares estimation model is established with the goal of minimizing the weighted sum of the quadratic forms of the observation residuals and the quadratic forms of the constraint residuals, and the optimal estimate of the inter-frequency deviation is obtained by solving the model.

[0021] The initial delay value is corrected using the optimal estimate.

[0022] Furthermore, step 3, which employs a mechanism that integrates RAIM and LSTM for early warning and recovery, includes:

[0023] When the observation quality is excellent, the LSTM time series prediction model is trained using dual-frequency pseudorange, carrier phase, Doppler, IMU data and observation quality indicators as inputs.

[0024] When the quality of the observations deteriorates, the LSTM time series prediction model generates initial predictions and constructs a multi-source input set that includes LSTM predictions, historical observations at the current frequency, valid observations at other frequencies, and IMU extrapolations.

[0025] The RAIM multi-source residual test is performed on the multi-source input set. After removing the gross error data sources whose residuals exceed the threshold one by one, the remaining data are solved by least squares to obtain the final recovered observations.

[0026] An early warning threshold is set based on the LSTM prediction error. When the prediction error exceeds the threshold and the proportion of gross errors exceeds the set ratio, an early warning signal is sent.

[0027] Furthermore, in step 4, the hybrid ionosphere delay processing strategy is as follows:

[0028] For dual-frequency observations, the deionization combination after correcting the inter-frequency deviation in step 2 is used;

[0029] For single-frequency observations, the ionospheric delay provided by the broadcast ionospheric model is used as the initial value, and this delay is incorporated into the positioning model as a parameter to be estimated. At the same time, the global ionospheric map (GIM) product is introduced as a non-difference, non-combination model with constraints.

[0030] Furthermore, in the non-difference, non-combination model for single-frequency observations, when the Global Ionospheric Map (GIM) product is used as a constraint, its constraint weights are dynamically weighted according to the elevation angle of the satellite corresponding to the observation: the higher the satellite elevation angle, the higher the assigned constraint weight.

[0031] Furthermore, in step 5, the adaptive decision-making model dynamically selects the combination mode, specifically including:

[0032] The number of available satellites, average carrier signal-to-noise ratio, inertial navigation position drift, rate of change of satellite visibility between epochs, and root mean square of positioning error in the preceding epoch are selected as core input features.

[0033] The quantitative input features are transformed into fuzzy sets through fuzzy logic, and the membership degree of each feature is calculated to form a membership vector.

[0034] The membership vector is input into a BP neural network trained with multi-scene samples, and the network outputs the probabilities of pure satellite navigation mode, GNSS / INS loose combination mode, and GNSS / INS tight combination mode.

[0035] Choose the pattern with the highest probability as the combination strategy for the current epoch.

[0036] Furthermore, when a warning signal is received from step 3, the adaptive decision model prioritizes increasing the inertial navigation weight ratio and shortens the transition time of the combined mode switching to 1 epoch.

[0037] On the other hand, the present invention provides a low-cost intelligent terminal satellite-inertial combined precision single-point positioning device, comprising:

[0038] The observation quality analysis module is used to acquire GNSS observation data and IMU data from low-cost smart terminals; based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation; the quality analysis is performed on the corrected GNSS observation data, and when abnormal or missing observation data is detected, the mechanism of fusion receiver autonomous integrity monitoring RAIM and long short-term memory network LSTM is used for early warning and recovery.

[0039] The combined solution decision module is used to construct an observation model for positioning solution based on the type of recovered observations using a hybrid ionospheric delay processing strategy; and dynamically selects the combination mode of GNSS and INS through an adaptive decision model that integrates fuzzy logic and neural networks.

[0040] The position parameter estimation module is used to dynamically construct the system state equation based on the selected combination mode and the corresponding observation model, combined with the IMU data, to estimate the position parameters using extended Kalman filtering, and then output the final positioning result after post-processing.

[0041] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method.

[0042] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method.

[0043] The beneficial effects of this invention are as follows:

[0044] This invention, through systematic technological innovation, brings significant benefits to low-cost smart terminals. Regarding positioning accuracy, a semi-parametric estimation model accurately corrects large numerical, slowly varying frequency deviations that traditional methods cannot handle. Combined with a hybrid ionospheric processing strategy, it effectively improves observation utilization and ionospheric delay correction accuracy, overcoming hardware performance limitations from the data source. In terms of robustness, a closed-loop processing mechanism fusing RAIM and LSTM is constructed, enabling proactive early warning and high-precision recovery of observation anomalies. Combined with a fuzzy-neural network-driven intelligent combination switching model, it significantly enhances positioning continuity and stability in complex dynamic scenarios. Regarding adaptability, all solutions are designed for low-cost hardware characteristics, achieving compatibility with various smart terminals and diverse application scenarios without hardware upgrades, thus achieving broad applicability. Furthermore, the integrated framework achieves synergistic effects across all technical aspects, reducing deployment complexity and computational consumption while ensuring high performance. Ultimately, this invention enables low-cost terminals to achieve near-professional-level high-precision, high-reliability positioning without relying on external infrastructure, significantly lowering the application threshold for high-precision positioning and possessing extremely high engineering application and market promotion value. Attached Figure Description

[0045] Figure 1 This is a flowchart of a low-cost intelligent terminal satellite-inertial combined precise single-point positioning method according to the present invention;

[0046] Figure 2 This is a schematic diagram illustrating the variation of the difference between the L1 pseudorange and L5 pseudorange of the Google Pixel 4 low-cost smart terminal G30 satellite between epochs.

[0047] Figure 3 The flowchart shows the RAIM-LSTM algorithm for early warning and recovery of outlier observations.

[0048] Figure 4 This is a schematic diagram of the fuzzy neural network adaptive switching decision. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0050] like Figure 1As shown, this invention breaks through the limitations of traditional technologies that "handle single problems in isolation," constructing a full-chain processing system of "deviation correction - data repair - ionospheric adaptation - combined optimization - integrated solution" to solve the problems of insufficient accuracy and poor robustness of precise point positioning (PPP) in low-cost smart terminals (such as IoT terminals, consumer smartphones, and low-cost drone navigation devices) due to low GNSS antenna gain and weak receiver chip processing capabilities. The specific methods for low-cost smart terminal satellite-inertial integrated precise point positioning provided by this invention include:

[0051] Step 1: Acquire GNSS observation data and IMU data from a low-cost smart terminal;

[0052] Step 2: Based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation;

[0053] Step 3: Perform quality analysis on the corrected GNSS observation data. When abnormal or missing observation data is detected, use the mechanism of autonomous integrity monitoring RAIM and long short-term memory network LSTM for early warning and recovery.

[0054] Step 4: Based on the type of recovered observations, construct an observation model for positioning calculation using a hybrid ionospheric delay processing strategy: For dual-frequency observations, an ionospheric de-combination model corrected for inter-frequency bias is used; for single-frequency observations, a non-difference, non-combination model is used with the broadcast ionospheric model as the initial value and the Global Ionospheric Map (GIM) product as the constraint.

[0055] Step 5: Based on the number of available satellites, average carrier signal-to-noise ratio, inertial navigation position drift, rate of change of satellite visibility between epochs, and historical positioning error, dynamically select the combination mode of GNSS and INS through an adaptive decision-making model that integrates fuzzy logic and neural networks.

[0056] Step 6: Based on the selected combination mode and the corresponding observation model, dynamically construct the system state equation using the IMU data, estimate the position parameters using extended Kalman filtering, and output the final positioning result after post-processing.

[0057] Furthermore, in step 1, the GNSS observation data and IMU data of the low-cost smart terminal are read; at the same time, the precise ephemeris, precise clock bias and other products required by PPP are read, and then the orbital coordinates and clock bias of the satellite corresponding to the GNSS observation value are calculated based on these products.

[0058] In step 2, addressing the core obstacle of low-cost smart terminals' dual-frequency pseudorange inter-frequency deviation (reaching thousands to tens of thousands of nanoseconds) and its slow change between epochs, an initial value of ionospheric delay is first calculated based on a broadcast ionospheric model or a Global Ionospheric Map (GIM) model. A semi-parametric estimation model is then constructed based on this value—defining the inter-frequency deviation as a non-modeling error that needs to be estimated separately. The expression for the inter-frequency deviation is derived through the dual-frequency pseudorange observation equation. Simultaneously, constraint equations are constructed to limit the maximum change in inter-frequency deviation between adjacent epochs, taking into account the characteristic of its gradual change between epochs. Finally, the compensated least squares method is used to solve the objective function, namely, "minimizing the weighted sum of the quadratic form of the observation residual and the quadratic form of the constraint residual," achieving accurate estimation of the inter-frequency deviation. The estimation results are then used to correct the original dual-frequency pseudorange observation values, providing a reliable data foundation for subsequent positioning calculations. Specifically: Considering the pseudorange inter-frequency deviation, the pseudorange observation equation is established as follows:

[0059] (1)

[0060] In the formula, and This indicates that a satellite has two frequencies. and pseudorange observations; The geometric distance between the terminal and the satellite; The speed of light; and These are the clock differences between the receiver and the satellite, respectively. For the terminal and satellite Ionospheric slack delay of frequency signals; For tropospheric delay between the terminal and the satellite; For two frequencies and Inter-frequency deviation; For unmodeled errors and noise, subscripts i and j correspond to different frequencies of a specific satellite. For measurement receivers, the inter-frequency offset is very small and can generally be modeled as a white noise parameter, estimated in PPP filtering. However, for some low-cost navigation terminals, the inter-frequency offset is large and changes slowly between epochs, such as... Figure 2 The epochal variation of the pseudo-range-frequency offset between L1 and L5 of the G30 satellite, used by the Google Pixel 4 low-cost smart terminal, is shown. Therefore, it can be estimated using a semi-parametric estimation method. Combining the two equations in equation (1), the expression for the frequency offset is obtained as follows:

[0061] (2)

[0062] The ionospheric delay can be obtained using a broadcast model or GIM product. Based on the above equation, the initial value of the inter-frequency deviation can be obtained. Considering its slowly changing characteristics, the following constraint equation is constructed:

[0063] (3)

[0064] in It is the inter-frequency offset vector of each satellite navigation system. Let C be the residual vector of the inter-frequency offset of the satellite navigation system, and let C represent the intermediate matrix. Based on the initial value of the inter-frequency offset, the residual equation is constructed according to equation (1):

[0065] (4)

[0066] The following objective function is established, and compensated least squares estimation is performed on it:

[0067] (5)

[0068] in, The observation noise matrix; Let T be the noise matrix of the inter-frequency deviation constraint equation, with the superscript T indicating transpose.

[0069] In step 3, due to the poor performance of the GNSS antenna and baseband chip in low-cost smart terminals, abnormal situations such as missing observations, carrier phase cycle slips, and pseudorange observation clock slips often occur in real-world scenarios. In order to ensure the quality of the observations, it is necessary to recover the abnormal observations. Traditional methods often use GNSS Doppler data or inertial measurement unit (IMU) data alone to recover the abnormal observations. Figure 3 As shown, this invention innovatively integrates the observation processing mechanism of RAIM and LSTM, and uses the quality status of GNSS observations from low-cost intelligent terminals as the core triggering basis to construct a closed-loop processing system of "pre-warning - in-process recovery - post-verification".

[0070] The overall triggering logic is divided into three states based on the quality of the observed values: when the carrier signal-to-noise ratio C / N0 ≥ 40dBHz and the epoch integrity ≥ 95%, it is a "high-quality state" and focuses on LSTM model training; when C / N0 < 30dBHz or the epoch integrity < 90%, it is a "deteriorating state" and initiates recovery and early warning; when it is in a "transitional state" of 30dBHz ≤ C / N0 < 40dBHz and 90% ≤ epoch integrity < 95%, it maintains LSTM prediction standby and quality monitoring.

[0071] Under optimal conditions, an LSTM time-series prediction model is constructed. The input consists of 12 features (2 dimensions each for dual-frequency pseudorange, carrier phase, and Doppler; 3 dimensions each for IMU three-axis acceleration and angular velocity, combined with the mean carrier signal-to-noise ratio and epoch missing rate). The model adopts an "input layer-hidden layer-output layer" structure: the input layer has 12 neurons, the hidden layer contains 32 LSTM units using the ReLU activation function, and the output layer corresponds to the predicted dual-frequency pseudorange and carrier phase values ​​for the next epoch. Model training employs an "offline pre-training + online fine-tuning" mode. The offline pre-training stage uses observation datasets from multiple scenarios, including urban areas, suburbs, and highways (each scenario has ≥100 samples). Taking dual-frequency pseudorange observations as an example, a mean squared error is constructed. Let be the loss function (where This is a reference value for dual-frequency pseudorange. The model is stopped and the base model is saved when the validation set MSE < 5m² (where N is the number of samples), and the predicted value is the dual-frequency pseudorange. During the online fine-tuning phase, the input and output layer parameters are fine-tuned using real-time data every 100 high-quality epochs to adapt to individual device differences. Simultaneously, the model calculates the mean prediction error using a sliding window (20 epochs). with standard deviation When three consecutive epochs satisfy When this occurs, it is identified as the beginning of an abnormal trend and the probability of abnormality is accumulated.

[0072] After entering a deteriorating state, taking pseudorange observation anomalies as an example, we first generate initial pseudorange predictions using an LSTM model, and then construct a multi-source input set. ,in For LSTM predictions, This represents the average pseudorange of the top 5 high-quality epochs at this frequency. The conversion results are for valid observations at other frequencies. The pseudorange is extrapolated from the IMU. Then, the RAIM multi-source residual test is initiated: pseudorange observation equations are constructed for each data source, and the sum of squared residuals is solved by simultaneously solving all equations. Then, eliminate individual equations one by one and calculate the sum of squared residuals. If the difference If the error exceeds 2.5 times the standard deviation of the observed values, the corresponding data source is considered an outlier and is discarded. The remaining data without outliers are then recalculated using least squares to obtain the final recovered pseudorange. Meanwhile, based on more than 1,000 sets of abnormal epoch data collected by the terminal, the 95th percentile of the LSTM prediction error is calculated as the initial warning threshold, and updated in real time through a sliding window of 60 epochs; when the prediction error exceeds the threshold and the gross error ratio is greater than 50%, a warning signal is sent to the combined solution module to increase the inertial navigation weight ratio in advance.

[0073] After recovery, the validity is confirmed by double verification: (1) Internal consistency verification is performed by comparing the recovered pseudorange. With subsequent high-quality epochal true pseudo-distance The deviation between them, if they satisfy If the result is positive, it is valid; (2) External correlation verification is performed by comparing the deviation between the location corresponding to the recovered pseudorange and the IMU extrapolation location. If the absolute value is less than the preset judgment threshold (e.g., 3m), it is further confirmed to be reliable. For valid samples, they are added to the LSTM online fine-tuning dataset; for invalid samples, they are marked as high-weight samples and the model is forced to learn. Model fine-tuning is started once every 10 recovery processes are completed to continuously optimize performance.

[0074] In step 4, due to the poor GNSS signal processing capabilities of low-cost smart terminals, their GNSS observations often exhibit a mixture of dual-frequency and single-frequency observations. Traditional dual-frequency combined processing strategies generally ignore single-frequency observations, while single-frequency processing strategies cannot effectively overcome the influence of ionospheric delay. Therefore, this invention adopts a strategy of mixed processing of dual-frequency ionospheric de-de-situation and ionospheric constraint. For dual-frequency observations, the inter-frequency deviation correction results obtained in the preceding steps are used to determine the linear combination coefficients based on the frequency characteristics of the dual-frequency signals. The ionospheric delay effect is offset by the ionospheric de-deflection combination operation, while retaining the core positioning information such as the geometric distance and clock difference between the terminal and the satellite. For single-frequency observations, the ionospheric delay provided by the broadcast ionospheric model is used as the initial value and incorporated into the positioning model as the parameter to be estimated. Then, the GIM product is introduced as a constraint condition, and an accuracy weighting model is designed based on the satellite elevation angle. The higher the satellite elevation angle, the shorter the path of its signal through the ionosphere, and the higher the accuracy of the GIM product in describing the satellite's ionospheric delay, corresponding to a higher constraint weight. The lower the satellite elevation angle, the lower the weight coefficient is. In practical applications, the weight is dynamically matched by dividing the elevation angle gradient interval to ensure that the GIM constraint effect is adapted to the actual accuracy, thereby improving the utilization rate of single-frequency observations and ensuring the reliability of ionospheric delay estimation. Specifically, for dual-frequency observations, taking pseudorange as an example, considering the influence of inter-frequency deviation, the dual-frequency ionospheric de-deflection combination of equation (1) is performed:

[0075] (6)

[0076] In the formula, For two frequencies and The pseudo-distance combined with ionospheric observations; This represents the unmodeled error and noise after combination. The inter-frequency bias in the equation is estimated in step 2. For single-frequency observations, an ionospherically constrained, non-differential, non-combined model is used:

[0077] (7)

[0078] In the formula Obtained using GIM products; The error between the estimated ionospheric oblique delay and the GIM reference value is weighted based on the elevation angle of the satellite launched from the observed value.

[0079] In step 5, to address the problem that traditional satellite-inertial navigation (GNSS / INS) handover relies on a single satellite quantity threshold and is difficult to adapt to complex scenarios, a fuzzy-neural network fusion technique is used for adaptive handover decision-making. The core is to construct a multi-dimensional intelligent decision-making model by fusing the fuzzy feature quantization capability of fuzzy logic with the complex mapping learning capability of BP neural networks. This model dynamically selects between pure satellite navigation, loose GNSS / INS, or tight combination modes. The specific process revolves around "feature preprocessing - fuzzification - neural network inference - mode decision - dynamic transition," as follows: Figure 4 As shown, it is linked in real time with other modules of the system.

[0080] First, core input features are selected and preprocessed. Five key features are extracted from the system's operating status and standardized to the [0,1] interval: the number of available satellites is the number of satellites stably tracked in the current epoch; the average carrier signal-to-noise ratio is the average of the signal-to-noise ratios of all available satellites; the IMU inertial drift is the average deviation between the IMU extrapolation and the GNSS true position over 10 seconds; and the rate of change of satellite visibility between epochs is determined by... calculate, The number of visible satellites at epoch t reflects the degree of dynamic environmental occlusion; the root mean square of the positioning error for the first three epochs is used to measure historical stability based on the Kalman filter position residual.

[0081] The process then proceeds to the fuzzy logic processing stage, where the five quantitative features are transformed into fuzzy sets and their membership degrees are calculated. Fuzzy sets are categorized according to document standards: available satellites ≥10 (excellent), 5-9 (medium), <5 (poor); average C / N0 ≥40dBHz (excellent), 30-40dBHz (medium), <30dBHz (poor); IMU position drift ≤1m / 10s (small), 1-5m / 10s (medium), >5m / 10s (large); satellite visibility change rate ≤10% (low), 10%-30% (medium), >30% (high); positioning error RMS ≤3m (excellent), 3-10m (medium), >10m (poor). A triangular membership function is used, with the vertex coordinates being the midpoint of each fuzzy interval. The membership degree of each feature to the three fuzzy sets is calculated, ultimately forming a 15-dimensional membership vector as input to the BP neural network.

[0082] The BP neural network modeling stage employs a three-layer fully connected structure: the input layer has 15 neurons corresponding to a 15-dimensional membership vector; the hidden layer has 10 neurons using the sigmoid activation function and introducing a momentum factor to accelerate convergence; and the output layer has 3 neurons using the softmax function to output the probabilities of pure satellite orientation, loose combination, and tight combination. Training samples cover typical scenarios such as urban canyons, high-speed movement, tree shading, and open areas, with ≥100 sets for each scenario. Each sample contains "15-dimensional input + optimal mode label" (expert annotation). During online inference, the membership vector of the current epoch is input into the network, and the network outputs the probability values ​​of the three modes.

[0083] The combined mode decision follows the "maximum probability priority" principle: the pure satellite navigation mode uses only GNSS observations, resulting in the lowest computational complexity; the loose combination fuses GNSS and IMU results in the location domain; and the tight combination directly corrects GNSS observations in the observation domain. When the new mode is inconsistent with the current mode, a linear weighted smooth transition is adopted: under normal circumstances, the switch is completed in 3-5 epochs; if a high-risk warning signal is received from the RAIM-LSTM module, the transition time is shortened to 1 epoch and the IMU weight is increased first.

[0084] This module needs to work in conjunction with other modules in the system: when receiving RAIM-LSTM early warning signals, it should force an increase in the probability weight of the combined mode; and it should send the mode decision results to the position parameter estimation module to guide it in constructing the system and observation equations.

[0085] In step 6, the position parameter estimation module uses the extended Kalman filter algorithm to set the process noise covariance based on the characteristics of the low-cost MEMS IMU and dynamically adjust the observation noise covariance according to the quality level of the observation. After filtering, the output is smoothed by a 5-epoch moving average filter and abnormal results with positioning errors exceeding 3 times the root mean square are removed. Finally, a stable and high-precision positioning result is output.

[0086] On the other hand, the present invention provides a low-cost intelligent terminal satellite-inertial navigation combined precision single-point positioning device, the various modules of which can realize the various steps of the aforementioned method, specifically including:

[0087] The observation quality analysis module is used to acquire GNSS observation data and IMU data from low-cost smart terminals; based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation; the quality analysis is performed on the corrected GNSS observation data, and when abnormal or missing observation data is detected, the mechanism of fusion receiver autonomous integrity monitoring RAIM and long short-term memory network LSTM is used for early warning and recovery.

[0088] The combined solution decision module is used to construct an observation model for positioning solution based on the type of recovered observations using a hybrid ionospheric delay processing strategy; and dynamically selects the combination mode of GNSS and INS through an adaptive decision model that integrates fuzzy logic and neural networks.

[0089] The position parameter estimation module is used to dynamically construct the system state equation based on the selected combination mode and the corresponding observation model, combined with the IMU data, to estimate the position parameters using extended Kalman filtering, and then output the final positioning result after post-processing.

[0090] Thirdly, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method.

[0091] Fourthly, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the aforementioned low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method.

[0092] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit 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 low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method, characterized in that, The method includes: Step 1: Acquire GNSS observation data and IMU data from a low-cost smart terminal; Step 2: Based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation; Step 3: Perform quality analysis on the corrected GNSS observation data. When abnormal or missing observation data is detected, use the mechanism of autonomous integrity monitoring RAIM and long short-term memory network LSTM for early warning and recovery. Step 4: Based on the type of the recovered observations, construct an observation model for location calculation using a hybrid ionospheric delay processing strategy; Step 5: Dynamically select the combination mode of GNSS and INS through an adaptive decision model that integrates fuzzy logic and neural networks; Step 6: Based on the selected combination mode and the corresponding observation model, dynamically construct the system state equation using the IMU data, estimate the position parameters using extended Kalman filtering, and output the final positioning result after post-processing.

2. The low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 1, characterized in that, Step 2 includes: Calculate the initial value of ionospheric delay based on the broadcast ionospheric model or the Global Ionospheric Map (GIM) product; Construct a pseudorange observation equation that defines the pseudorange-frequency deviation as a non-modeling error that needs to be estimated separately; The expression for the inter-frequency deviation is derived from the pseudorange observation equation to obtain its initial value; Construct constraint equations to limit the maximum change in frequency deviation between adjacent epochs; A compensated least squares estimation model is established with the goal of minimizing the weighted sum of the quadratic forms of the observation residuals and the quadratic forms of the constraint residuals, and the optimal estimate of the inter-frequency deviation is obtained by solving the model. The initial delay value is corrected using the optimal estimate.

3. The low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 1, characterized in that, Step 3, which employs a mechanism that integrates RAIM and LSTM for early warning and recovery, includes: When the observation quality is excellent, the LSTM time series prediction model is trained using dual-frequency pseudorange, carrier phase, Doppler, IMU data and observation quality indicators as inputs. When the quality of the observations deteriorates, the LSTM time series prediction model generates initial predictions and constructs a multi-source input set that includes LSTM predictions, historical observations at the current frequency, valid observations at other frequencies, and IMU extrapolations. The RAIM multi-source residual test is performed on the multi-source input set. After removing the gross error data sources whose residuals exceed the threshold one by one, the remaining data are solved by least squares to obtain the final recovered observations. An early warning threshold is set based on the LSTM prediction error. When the prediction error exceeds the threshold and the proportion of gross errors exceeds the set ratio, an early warning signal is sent.

4. The low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 1, characterized in that, In step 4, the hybrid ionosphere delay processing strategy is as follows: For dual-frequency observations, the deionization combination after correcting the inter-frequency deviation in step 2 is used; For single-frequency observations, the ionospheric delay provided by the broadcast ionospheric model is used as the initial value, and this delay is incorporated into the positioning model as a parameter to be estimated. At the same time, the global ionospheric map (GIM) product is introduced as a non-difference, non-combination model with constraints.

5. A low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 4, characterized in that, In the non-difference, non-combination model for single-frequency observations, when the Global Ionospheric Map (GIM) product is used as a constraint, its constraint weights are dynamically weighted according to the elevation angle of the satellite corresponding to the observation: the higher the satellite elevation angle, the higher the assigned constraint weight.

6. The low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 1, characterized in that, In step 5, the adaptive decision model dynamically selects the combination mode, specifically including: The number of available satellites, average carrier signal-to-noise ratio, inertial navigation position drift, rate of change of satellite visibility between epochs, and root mean square of positioning error in the preceding epoch are selected as core input features. The quantitative input features are transformed into fuzzy sets through fuzzy logic, and the membership degree of each feature is calculated to form a membership vector. The membership vector is input into a BP neural network trained with multi-scene samples, and the network outputs the probabilities of pure satellite navigation mode, GNSS / INS loose combination mode, and GNSS / INS tight combination mode. Choose the pattern with the highest probability as the combination strategy for the current epoch.

7. A low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to claim 6, characterized in that, When a warning signal is received from step 3, the adaptive decision model prioritizes increasing the inertial navigation weight ratio and shortens the transition time of the combined mode switching to 1 epoch.

8. A low-cost intelligent terminal satellite-inertial navigation combined precision single-point positioning device, characterized in that, include: The observation quality analysis module is used to acquire GNSS observation data and IMU data from low-cost smart terminals; Based on the GNSS observation data, a semi-parametric estimation method is used to estimate and correct the pseudo-range-frequency deviation; The quality of the corrected GNSS observation data is analyzed. When abnormal or missing observation data is detected, the mechanism of autonomous integrity monitoring RAIM and long short-term memory network LSTM is used for early warning and recovery. The combined solution decision module is used to construct an observation model for positioning solution based on the type of recovered observations using a hybrid ionospheric delay processing strategy; and dynamically selects the combination mode of GNSS and INS through an adaptive decision model that integrates fuzzy logic and neural networks. The position parameter estimation module is used to dynamically construct the system state equation based on the selected combination mode and the corresponding observation model, combined with the IMU data, to estimate the position parameters using extended Kalman filtering, and then output the final positioning result after post-processing.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the low-cost intelligent terminal satellite-inertial navigation combined precise single-point positioning method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions that, when executed by a processor, enable the processor to implement the low-cost intelligent terminal satellite-inertial combination precise single-point positioning method as described in any one of claims 1-7.

Images